From 4a2fef25522b90f1572c8b6e89320581e17c8df3 Mon Sep 17 00:00:00 2001
From: Chenhan Yu <chenhany@nvidia.com>
Date: Wed, 15 Apr 2026 12:11:00 -0700
Subject: [PATCH 1/6] fix: remove empty environment key causing nemo_run
 ListParseError

nemo_run's CLI parser fails when `environment:` is present but has no
value (parsed as None, not a valid List[Dict[str, str]]).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Chenhan Yu <chenhany@nvidia.com>
---
 tools/launcher/examples/Qwen/Qwen3-8B/hf_online_dflash.yaml | 1 -
 1 file changed, 1 deletion(-)

diff --git a/tools/launcher/examples/Qwen/Qwen3-8B/hf_online_dflash.yaml b/tools/launcher/examples/Qwen/Qwen3-8B/hf_online_dflash.yaml
index e255f51dc32..739933dd891 100644
--- a/tools/launcher/examples/Qwen/Qwen3-8B/hf_online_dflash.yaml
+++ b/tools/launcher/examples/Qwen/Qwen3-8B/hf_online_dflash.yaml
@@ -36,7 +36,6 @@ pipeline:
       - dflash.dflash_loss_decay_factor=7
       - dflash.dflash_mask_token_id=151669
       - dflash.dflash_architecture_config.num_hidden_layers=5
-    environment:
     slurm_config:
       _factory_: "slurm_factory"
       nodes: 1

From 728f33a0dc6b0f82b788f3358c2c20a412d9e637 Mon Sep 17 00:00:00 2001
From: Chenhan Yu <chenhany@nvidia.com>
Date: Fri, 17 Apr 2026 09:12:33 -0700
Subject: [PATCH 2/6] fix: vLLM smoke test regression check + query.sh server
 liveness

- vllm_smoke_test.sh: parse acceptance length from vLLM log instead of
  missing Prometheus metric; capture server output to temp file
- vllm_smoke_test.sh: pip install pandas for broken nightly container
- query.sh: detect server death + 600s startup timeout (prevents
  infinite polling when vLLM crashes, wasting GPU hours)
- query.sh: pip install pandas for broken nightly container
- hf_online_dflash.yaml: add regression baselines from B200 100K run,
  add MAX_FINAL_LOSS/MIN_FINAL_ACC/MIN_ACCEPTANCE_LENGTH thresholds

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Chenhan Yu <chenhany@nvidia.com>
---
 .../common/specdec/vllm_smoke_test.sh         | 35 ++++++++++++++++---
 tools/launcher/common/vllm/query.sh           | 20 +++++++++--
 .../Qwen/Qwen3-8B/hf_online_dflash.yaml       | 25 +++++++++++--
 3 files changed, 72 insertions(+), 8 deletions(-)

diff --git a/tools/launcher/common/specdec/vllm_smoke_test.sh b/tools/launcher/common/specdec/vllm_smoke_test.sh
index 1e508cb1169..4b9d5a63b4f 100644
--- a/tools/launcher/common/specdec/vllm_smoke_test.sh
+++ b/tools/launcher/common/specdec/vllm_smoke_test.sh
@@ -32,7 +32,10 @@
 SCRIPT_DIR="$(dirname "$(readlink -f "$0")")"
 source ${SCRIPT_DIR}/../service_utils.sh 2>/dev/null || true
 
-cleanup() { kill $SERVER_PID 2>/dev/null; sleep 2; kill -9 $SERVER_PID 2>/dev/null; }
+# Ensure pandas is available (missing in some vLLM nightly builds)
+pip install pandas 2>/dev/null || true
+
+cleanup() { kill $SERVER_PID 2>/dev/null; sleep 2; kill -9 $SERVER_PID 2>/dev/null; rm -f "${VLLM_LOG:-}" 2>/dev/null; }
 trap cleanup EXIT
 
 MODEL=${HF_MODEL_CKPT}
@@ -72,7 +75,8 @@ if [ "${DISABLE_PREFIX_CACHING:-}" = "1" ]; then
     OPTIONAL_ARGS="${OPTIONAL_ARGS} --no-enable-prefix-caching"
 fi
 
-# Start vLLM server
+# Start vLLM server (capture output for regression check parsing)
+VLLM_LOG=$(mktemp /tmp/vllm_server_XXXXXX.log)
 if [ -n "$SPEC_CONFIG" ]; then
     vllm serve ${MODEL} \
         --speculative-config "${SPEC_CONFIG}" \
@@ -80,14 +84,14 @@ if [ -n "$SPEC_CONFIG" ]; then
         --tensor-parallel-size ${TP} \
         --port ${PORT} \
         ${OPTIONAL_ARGS} \
-        &
+        > >(tee -a "$VLLM_LOG") 2>&1 &
 else
     vllm serve ${MODEL} \
         --max-num-batched-tokens 32768 \
         --tensor-parallel-size ${TP} \
         --port ${PORT} \
         ${OPTIONAL_ARGS} \
-        &
+        > >(tee -a "$VLLM_LOG") 2>&1 &
 fi
 SERVER_PID=$!
 
@@ -168,4 +172,27 @@ if [ $FAIL -gt 0 ]; then
     exit 1
 fi
 
+# Regression check: minimum acceptance length for speculative decoding
+if [ -n "${MIN_ACCEPTANCE_LENGTH:-}" ]; then
+    # Parse mean acceptance length from vLLM's SpecDecoding metrics log.
+    # vLLM logs: "SpecDecoding metrics: Mean acceptance length: X.XX, ..."
+    # Take the last reported value (most accurate, covers all prompts).
+    AVG_ACCEPT=$(grep -oP 'Mean acceptance length: \K[0-9.]+' "$VLLM_LOG" 2>/dev/null | tail -1 || true)
+    if [ -n "$AVG_ACCEPT" ]; then
+        echo ""
+        echo "=== Acceptance Length Regression Check ==="
+        echo "  Mean acceptance length: ${AVG_ACCEPT}"
+        echo "  Threshold: ${MIN_ACCEPTANCE_LENGTH}"
+        PASS_CHECK=$(python3 -c "print('yes' if float('${AVG_ACCEPT}') >= float('${MIN_ACCEPTANCE_LENGTH}') else 'no')")
+        if [ "$PASS_CHECK" = "yes" ]; then
+            echo "  PASS: ${AVG_ACCEPT} >= ${MIN_ACCEPTANCE_LENGTH}"
+        else
+            echo "  REGRESSION: ${AVG_ACCEPT} < ${MIN_ACCEPTANCE_LENGTH}"
+            exit 1
+        fi
+    else
+        echo "WARNING: Could not parse acceptance length from vLLM log, skipping regression check"
+    fi
+fi
+
 echo "Done"
diff --git a/tools/launcher/common/vllm/query.sh b/tools/launcher/common/vllm/query.sh
index 4ce6ded1965..d1513623c34 100755
--- a/tools/launcher/common/vllm/query.sh
+++ b/tools/launcher/common/vllm/query.sh
@@ -58,6 +58,9 @@ source ${SCRIPT_DIR}/../service_utils.sh
 #     gpus_per_node: 4
 ###################################################################################################
 
+# Ensure pandas is available (missing in some vLLM nightly builds)
+pip install pandas 2>/dev/null || true
+
 export OPENAI_API_KEY="token-abc123"
 
 if [ -z ${SLURM_ARRAY_TASK_ID} ]; then
@@ -108,13 +111,26 @@ SERVER_PID=$!
 
 # Wait for server to start up by polling the health endpoint
 echo "Waiting for server to start..."
+MAX_WAIT=${VLLM_STARTUP_TIMEOUT:-600}
+WAITED=0
 while true; do
+    if ! kill -0 $SERVER_PID 2>/dev/null; then
+        echo "ERROR: vLLM server process died during startup"
+        wait $SERVER_PID 2>/dev/null
+        exit 1
+    fi
     response=$(curl -s -o /dev/null -w "%{http_code}" "http://$(hostname -f):8000/health" || true)
     if [ "$response" -eq 200 ]; then
-        echo "Server is up!"
+        echo "Server is up! (waited ${WAITED}s)"
         break
     fi
-    echo "Server not ready yet, retrying in 10 seconds..."
+    WAITED=$((WAITED + 10))
+    if [ $WAITED -ge $MAX_WAIT ]; then
+        echo "ERROR: vLLM server failed to start within ${MAX_WAIT}s"
+        kill $SERVER_PID 2>/dev/null
+        exit 1
+    fi
+    echo "Server not ready yet (${WAITED}/${MAX_WAIT}s), retrying in 10 seconds..."
     sleep 10
 done
 
diff --git a/tools/launcher/examples/Qwen/Qwen3-8B/hf_online_dflash.yaml b/tools/launcher/examples/Qwen/Qwen3-8B/hf_online_dflash.yaml
index 739933dd891..7c7f2a959dc 100644
--- a/tools/launcher/examples/Qwen/Qwen3-8B/hf_online_dflash.yaml
+++ b/tools/launcher/examples/Qwen/Qwen3-8B/hf_online_dflash.yaml
@@ -5,6 +5,21 @@
 #   task_1: vLLM smoke test with DFlash speculative decoding
 #   task_2: MT-Bench per-category HF AR evaluation (1 GPU)
 #
+# Convergence baseline (8x B200, batch_size=1, seq_len=4096, 5-layer draft, block_size=16):
+#   100K samples, 1 epoch (~12,500 steps)
+#   Step   100 (epoch 0.01): loss=8.900  acc=0.029
+#   Step  1000 (epoch 0.08): loss=5.845  acc=0.096
+#   Step  2500 (epoch 0.20): loss=4.981  acc=0.138
+#   Step  5000 (epoch 0.40): loss=4.383  acc=0.176
+#   Step  7500 (epoch 0.60): loss=4.040  acc=0.196
+#   Step 10000 (epoch 0.80): loss=3.900  acc=0.210
+#   Step 12500 (epoch 1.00): loss=3.821  acc=0.200
+#   Average train_loss=4.493, training time=5094s
+#
+# Regression criteria (set via environment):
+#   MAX_FINAL_LOSS: final loss must be below this (default: 5.0)
+#   MIN_FINAL_ACC:  final accuracy must be above this (default: 0.15)
+#
 # Reference: "DFlash: Block Diffusion for Flash Speculative Decoding" (arXiv:2602.06036)
 #
 # Usage:
@@ -22,13 +37,14 @@ pipeline:
     args:
       - --config modules/Model-Optimizer/modelopt_recipes/general/speculative_decoding/dflash.yaml
       - model.model_name_or_path=<<global_vars.hf_model>>
-      - data.data_path=/hf-local/modelopt/Speculative-Decoding-Dataset-v1-Qwen3-8B/sample-1K-openai.jsonl
+      - data.data_path=/hf-local/modelopt/Speculative-Decoding-Dataset-v1-Qwen3-8B/sample-100K-openai.jsonl
       - data.chat_template=examples/Qwen/Qwen3-8B/chat_template_train.jinja
       - training.output_dir=/scratchspace/dflash_bs16
+      - training.per_device_train_batch_size=1
       - training.num_train_epochs=1
       - training.training_seq_len=4096
       - training.save_steps=5000
-      - training.logging_steps=1000
+      - training.logging_steps=100
       - training.disable_tqdm=true
       - training.answer_only_loss=true
       - dflash.dflash_block_size=16
@@ -36,6 +52,9 @@ pipeline:
       - dflash.dflash_loss_decay_factor=7
       - dflash.dflash_mask_token_id=151669
       - dflash.dflash_architecture_config.num_hidden_layers=5
+    environment:
+      - MAX_FINAL_LOSS: "5.0"
+      - MIN_FINAL_ACC: "0.15"
     slurm_config:
       _factory_: "slurm_factory"
       nodes: 1
@@ -50,8 +69,10 @@ pipeline:
       - DRAFT_CKPT_DIR: /scratchspace/dflash_bs16
       - SPEC_METHOD: "dflash"
       - NUM_SPEC_TOKENS: "7"
+      - MIN_ACCEPTANCE_LENGTH: "1.4"
     slurm_config:
       _factory_: "slurm_factory"
+      container: "vllm/vllm-openai:nightly"
       nodes: 1
       ntasks_per_node: 1
       gpus_per_node: 1

From 210d7ae3d5671a551e96353f963a644787639ea3 Mon Sep 17 00:00:00 2001
From: Chenhan Yu <chenhany@nvidia.com>
Date: Fri, 17 Apr 2026 15:13:02 -0700
Subject: [PATCH 3/6] update: Qwen3-8B PTQ and EAGLE3 to 8 GPUs

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Chenhan Yu <chenhany@nvidia.com>
---
 .../Qwen/Qwen3-8B/hf_offline_eagle3.yaml      | 22 +++++++++----------
 .../Qwen/Qwen3-8B/megatron_lm_ptq.yaml        |  8 +++----
 2 files changed, 15 insertions(+), 15 deletions(-)

diff --git a/tools/launcher/examples/Qwen/Qwen3-8B/hf_offline_eagle3.yaml b/tools/launcher/examples/Qwen/Qwen3-8B/hf_offline_eagle3.yaml
index ae2c1e957cf..24068c4bb3e 100644
--- a/tools/launcher/examples/Qwen/Qwen3-8B/hf_offline_eagle3.yaml
+++ b/tools/launcher/examples/Qwen/Qwen3-8B/hf_offline_eagle3.yaml
@@ -27,8 +27,8 @@ pipeline:
     script: common/tensorrt_llm/query.sh
     args:
       - --model <<global_vars.hf_model>>
-      - --tp_size 4
-      - --ep_size 4
+      - --tp_size 8
+      - --ep_size 8
       - --max_num_tokens 32000
       - --port 8000
       - --host 0.0.0.0
@@ -41,8 +41,8 @@ pipeline:
     slurm_config:
       _factory_: "slurm_factory"
       nodes: 1
-      ntasks_per_node: 4
-      gpus_per_node: 4
+      ntasks_per_node: 8
+      gpus_per_node: 8
       container: nvcr.io/nvidia/tensorrt-llm/release:1.2.0
 
   # Step 2: Dump hidden states from target model
@@ -52,15 +52,15 @@ pipeline:
       - --input-data /scratchspace/data
       - --output-dir /scratchspace/offline_hidden_states
       - --max-seq-len 8192
-      - --tp 4
-      - --moe-ep 4
+      - --tp 8
+      - --moe-ep 8
     environment:
       - HF_MODEL_CKPT: <<global_vars.hf_model>>
     slurm_config:
       _factory_: "slurm_factory"
       nodes: 1
-      ntasks_per_node: 4
-      gpus_per_node: 4
+      ntasks_per_node: 8
+      gpus_per_node: 8
       container: nvcr.io/nvidia/tensorrt-llm/release:1.2.0
 
   # Step 3: Train EAGLE3 draft head (offline, single task)
@@ -78,7 +78,7 @@ pipeline:
       _factory_: "slurm_factory"
       nodes: 1
       ntasks_per_node: 1
-      gpus_per_node: 4
+      gpus_per_node: 8
       container: nvcr.io/nvidia/tensorrt-llm/release:1.2.0
 
   # Step 4: Benchmark speculative decoding (VLLM backend)
@@ -89,7 +89,7 @@ pipeline:
       - --draft_length 3
       - --output_length 4096
       - --engine VLLM
-      - --tp_size 4
+      - --tp_size 8
       - --ep_size 1
       - --speculative_algorithm EAGLE3
       - --mtbench /hf-local/HuggingFaceH4/mt_bench_prompts/raw/question.jsonl
@@ -100,5 +100,5 @@ pipeline:
       _factory_: "slurm_factory"
       nodes: 1
       ntasks_per_node: 1
-      gpus_per_node: 4
+      gpus_per_node: 8
       container: vllm/vllm-openai:latest
diff --git a/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml b/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml
index a67d0bceb07..c9f667e31b6 100644
--- a/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml
+++ b/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml
@@ -1,4 +1,4 @@
-# Qwen3-8B NVFP4 quantization (4 GPUs, for Slurm clusters).
+# Qwen3-8B NVFP4 quantization (8 GPUs, for Slurm clusters).
 #
 # Uses MegatronLMQuantizeTask with typed config — see common/megatron_lm/quantize/task.py
 # for all available fields.
@@ -19,7 +19,7 @@ pipeline:
     config:
       model: Qwen/Qwen3-8B
       quant_cfg: NVFP4_DEFAULT_CFG
-      tp: 4
+      tp: 8
       calib_dataset: abisee/cnn_dailymail
       calib_size: 32
       mmlu_dataset: cais/mmlu
@@ -27,5 +27,5 @@ pipeline:
     slurm_config:
       _factory_: "slurm_factory"
       nodes: 1
-      ntasks_per_node: 4
-      gpus_per_node: 4
+      ntasks_per_node: 8
+      gpus_per_node: 8

From 4c209fdc4afc094239eb83a703df25b09e5dc5f5 Mon Sep 17 00:00:00 2001
From: Chenhan Yu <chenhany@nvidia.com>
Date: Fri, 17 Apr 2026 16:26:16 -0700
Subject: [PATCH 4/6] update: Qwen3-8B PTQ with FP8+NVFP4 and MMLU threshold
 0.68

- Add FP8_DEFAULT_CFG as task_1 alongside NVFP4 task_0
- Pass MMLU_LOWER_BOUND from typed config to quantize.sh
- Raise MMLU threshold to 0.68 (matches quantize_resume test)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Chenhan Yu <chenhany@nvidia.com>
---
 .../common/megatron_lm/quantize/quantize.sh   |  2 +-
 .../common/megatron_lm/quantize/task.py       |  1 +
 .../Qwen/Qwen3-8B/megatron_lm_ptq.yaml        | 24 +++++++++++++++++--
 3 files changed, 24 insertions(+), 3 deletions(-)

diff --git a/tools/launcher/common/megatron_lm/quantize/quantize.sh b/tools/launcher/common/megatron_lm/quantize/quantize.sh
index 6e4d21b9945..7d09e83bb5c 100755
--- a/tools/launcher/common/megatron_lm/quantize/quantize.sh
+++ b/tools/launcher/common/megatron_lm/quantize/quantize.sh
@@ -38,7 +38,7 @@ EXPORT_EXE="bash modules/Megatron-LM/examples/post_training/modelopt/export.sh"
 export MLM_EXTRA_ARGS=${@}
 ${QUANTIZE_EXE} ${MLM_MODEL_CFG} ${QUANT_CFG}
 
-export MLM_EXTRA_ARGS="--mmlu-dataset ${MMLU_DATASET:-/hf-local/cais/mmlu} --fraction 0.01 --lower-bound 0.38 --disable-tqdm"
+export MLM_EXTRA_ARGS="--mmlu-dataset ${MMLU_DATASET:-/hf-local/cais/mmlu} --fraction 0.01 --lower-bound ${MMLU_LOWER_BOUND:-0.38} --disable-tqdm"
 MLM_MODEL_CKPT=${MLM_MODEL_SAVE} ${MMLU_EXE} ${MLM_MODEL_CFG}
 
 ###################################################################################################
diff --git a/tools/launcher/common/megatron_lm/quantize/task.py b/tools/launcher/common/megatron_lm/quantize/task.py
index 4ba10030e61..313a238556c 100644
--- a/tools/launcher/common/megatron_lm/quantize/task.py
+++ b/tools/launcher/common/megatron_lm/quantize/task.py
@@ -102,4 +102,5 @@ def __post_init__(self):
                 {"HF_MODEL_CKPT": f"{c.hf_local}{c.model}"},
                 {"MMLU_DATASET": f"{c.hf_local}{c.mmlu_dataset}"},
                 {"TP": str(c.tp)},
+                {"MMLU_LOWER_BOUND": str(c.mmlu_lower_bound)},
             ]
diff --git a/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml b/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml
index c9f667e31b6..08d4dfccf2b 100644
--- a/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml
+++ b/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml
@@ -1,4 +1,6 @@
-# Qwen3-8B NVFP4 quantization (8 GPUs, for Slurm clusters).
+# Qwen3-8B PTQ quantization (8 GPUs, for Slurm clusters).
+#
+# 2-step pipeline: NVFP4 then FP8, each followed by MMLU evaluation.
 #
 # Uses MegatronLMQuantizeTask with typed config — see common/megatron_lm/quantize/task.py
 # for all available fields.
@@ -8,7 +10,7 @@
 #
 # For single-GPU local Docker, use megatron_lm_ptq_local.yaml instead.
 
-job_name: Qwen3-8B_NVFP4_DEFAULT_CFG
+job_name: Qwen3-8B_PTQ
 pipeline:
   skip: false
   allow_to_fail: false
@@ -23,6 +25,24 @@ pipeline:
       calib_dataset: abisee/cnn_dailymail
       calib_size: 32
       mmlu_dataset: cais/mmlu
+      mmlu_lower_bound: 0.68
+      hf_local: /hf-local/
+    slurm_config:
+      _factory_: "slurm_factory"
+      nodes: 1
+      ntasks_per_node: 8
+      gpus_per_node: 8
+
+  task_1:
+    _target_: common.megatron_lm.quantize.task.MegatronLMQuantizeTask
+    config:
+      model: Qwen/Qwen3-8B
+      quant_cfg: FP8_DEFAULT_CFG
+      tp: 8
+      calib_dataset: abisee/cnn_dailymail
+      calib_size: 32
+      mmlu_dataset: cais/mmlu
+      mmlu_lower_bound: 0.68
       hf_local: /hf-local/
     slurm_config:
       _factory_: "slurm_factory"

From 65eca0f32f2852475d620fb0c3c5050992627317 Mon Sep 17 00:00:00 2001
From: Chenhan Yu <chenhany@nvidia.com>
Date: Fri, 17 Apr 2026 17:03:35 -0700
Subject: [PATCH 5/6] add: Qwen3-30B-A3B PTQ example + EP/PP support in
 quantize task

- New launcher example: Qwen3-30B-A3B FP8+NVFP4 with MMLU >= 0.75
- Add pp, ep, etp, extra_args fields to MegatronLMQuantizeConfig
- Pass TP/PP/EP/ETP env vars to quantize.sh for both quantize and MMLU

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Chenhan Yu <chenhany@nvidia.com>
---
 .../common/megatron_lm/quantize/quantize.sh   |  4 +-
 .../common/megatron_lm/quantize/task.py       | 12 +++-
 .../Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml   | 55 +++++++++++++++++++
 3 files changed, 68 insertions(+), 3 deletions(-)
 create mode 100644 tools/launcher/examples/Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml

diff --git a/tools/launcher/common/megatron_lm/quantize/quantize.sh b/tools/launcher/common/megatron_lm/quantize/quantize.sh
index 7d09e83bb5c..678402b105e 100755
--- a/tools/launcher/common/megatron_lm/quantize/quantize.sh
+++ b/tools/launcher/common/megatron_lm/quantize/quantize.sh
@@ -36,10 +36,10 @@ CONVERT_EXE="bash modules/Megatron-LM/examples/post_training/modelopt/convert.sh
 EXPORT_EXE="bash modules/Megatron-LM/examples/post_training/modelopt/export.sh"
 
 export MLM_EXTRA_ARGS=${@}
-${QUANTIZE_EXE} ${MLM_MODEL_CFG} ${QUANT_CFG}
+TP=${TP:-1} PP=${PP:-1} EP=${EP:-1} ETP=${ETP:-1} ${QUANTIZE_EXE} ${MLM_MODEL_CFG} ${QUANT_CFG}
 
 export MLM_EXTRA_ARGS="--mmlu-dataset ${MMLU_DATASET:-/hf-local/cais/mmlu} --fraction 0.01 --lower-bound ${MMLU_LOWER_BOUND:-0.38} --disable-tqdm"
-MLM_MODEL_CKPT=${MLM_MODEL_SAVE} ${MMLU_EXE} ${MLM_MODEL_CFG}
+TP=${TP:-1} PP=${PP:-1} EP=${EP:-1} ETP=${ETP:-1} MLM_MODEL_CKPT=${MLM_MODEL_SAVE} ${MMLU_EXE} ${MLM_MODEL_CFG}
 
 ###################################################################################################
 
diff --git a/tools/launcher/common/megatron_lm/quantize/task.py b/tools/launcher/common/megatron_lm/quantize/task.py
index 313a238556c..95833fe3960 100644
--- a/tools/launcher/common/megatron_lm/quantize/task.py
+++ b/tools/launcher/common/megatron_lm/quantize/task.py
@@ -66,6 +66,10 @@ class MegatronLMQuantizeConfig:
     model: str = "Qwen/Qwen3-8B"
     quant_cfg: str = "NVFP4_DEFAULT_CFG"
     tp: int = 4
+    pp: int = 1
+    ep: int = 1
+    etp: int = 1
+    extra_args: str = ""
     calib_dataset: str = "abisee/cnn_dailymail"
     calib_size: int = 32
     mmlu_dataset: str = "cais/mmlu"
@@ -92,15 +96,21 @@ def __post_init__(self):
         if self.config is not None:
             c = self.config
             self.script = self.script or "common/megatron_lm/quantize/quantize.sh"
-            self.args = [
+            args = [
                 f"--calib-dataset-path-or-name {c.hf_local}{c.calib_dataset}",
                 f"--calib-size {c.calib_size}",
             ]
+            if c.extra_args:
+                args.append(c.extra_args)
+            self.args = args
             self.environment = [
                 {"MLM_MODEL_CFG": c.model},
                 {"QUANT_CFG": c.quant_cfg},
                 {"HF_MODEL_CKPT": f"{c.hf_local}{c.model}"},
                 {"MMLU_DATASET": f"{c.hf_local}{c.mmlu_dataset}"},
                 {"TP": str(c.tp)},
+                {"PP": str(c.pp)},
+                {"EP": str(c.ep)},
+                {"ETP": str(c.etp)},
                 {"MMLU_LOWER_BOUND": str(c.mmlu_lower_bound)},
             ]
diff --git a/tools/launcher/examples/Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml b/tools/launcher/examples/Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml
new file mode 100644
index 00000000000..7ba0566f254
--- /dev/null
+++ b/tools/launcher/examples/Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml
@@ -0,0 +1,55 @@
+# Qwen3-30B-A3B PTQ quantization (8 GPUs, MoE model).
+#
+# 2-step pipeline: NVFP4 then FP8, each followed by MMLU evaluation.
+# MMLU uses EP for expert parallelism.
+#
+# Usage:
+#   uv run launch.py --yaml examples/Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml --yes
+
+job_name: Qwen3-30B-A3B_PTQ
+pipeline:
+  skip: false
+  allow_to_fail: false
+  note:
+
+  task_0:
+    _target_: common.megatron_lm.quantize.task.MegatronLMQuantizeTask
+    config:
+      model: Qwen/Qwen3-30B-A3B
+      quant_cfg: NVFP4_DEFAULT_CFG
+      tp: 1
+      pp: 1
+      ep: 8
+      etp: 1
+      extra_args: ""
+      calib_dataset: abisee/cnn_dailymail
+      calib_size: 32
+      mmlu_dataset: cais/mmlu
+      mmlu_lower_bound: 0.75
+      hf_local: /hf-local/
+    slurm_config:
+      _factory_: "slurm_factory"
+      nodes: 1
+      ntasks_per_node: 8
+      gpus_per_node: 8
+
+  task_1:
+    _target_: common.megatron_lm.quantize.task.MegatronLMQuantizeTask
+    config:
+      model: Qwen/Qwen3-30B-A3B
+      quant_cfg: FP8_DEFAULT_CFG
+      tp: 1
+      pp: 1
+      ep: 8
+      etp: 1
+      extra_args: ""
+      calib_dataset: abisee/cnn_dailymail
+      calib_size: 32
+      mmlu_dataset: cais/mmlu
+      mmlu_lower_bound: 0.75
+      hf_local: /hf-local/
+    slurm_config:
+      _factory_: "slurm_factory"
+      nodes: 1
+      ntasks_per_node: 8
+      gpus_per_node: 8

From e8f52e553a28aacd81bddf076600b56c6a353edb Mon Sep 17 00:00:00 2001
From: Chenhan Yu <chenhany@nvidia.com>
Date: Fri, 17 Apr 2026 17:45:51 -0700
Subject: [PATCH 6/6] add: export + TRT-LLM eval in PTQ pipeline, fix empty
 extra_args

- quantize.sh: always export after MMLU (auto-detect GPUs for PP)
- Add common/tensorrt_llm/eval.sh for TRT-LLM MMLU evaluation
- Qwen3-8B PTQ: add task_2 for TRT-LLM eval on exported checkpoints
- Fix empty extra_args causing nemo_run parse error

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Signed-off-by: Chenhan Yu <chenhany@nvidia.com>
---
 .../common/megatron_lm/quantize/quantize.sh   |   8 +
 tools/launcher/common/tensorrt_llm/eval.sh    |  59 ++++
 .../tensorrt_llm/extra_llm_api_options.yaml   |  52 +++
 .../Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml   |   2 -
 .../Qwen/Qwen3-8B/megatron_lm_ptq.yaml        |  18 +-
 uv.lock                                       | 310 +++++++++++++++---
 6 files changed, 402 insertions(+), 47 deletions(-)
 create mode 100644 tools/launcher/common/tensorrt_llm/eval.sh
 create mode 100644 tools/launcher/common/tensorrt_llm/extra_llm_api_options.yaml

diff --git a/tools/launcher/common/megatron_lm/quantize/quantize.sh b/tools/launcher/common/megatron_lm/quantize/quantize.sh
index 678402b105e..1bb0d60e80d 100755
--- a/tools/launcher/common/megatron_lm/quantize/quantize.sh
+++ b/tools/launcher/common/megatron_lm/quantize/quantize.sh
@@ -41,6 +41,14 @@ TP=${TP:-1} PP=${PP:-1} EP=${EP:-1} ETP=${ETP:-1} ${QUANTIZE_EXE} ${MLM_MODEL_CF
 export MLM_EXTRA_ARGS="--mmlu-dataset ${MMLU_DATASET:-/hf-local/cais/mmlu} --fraction 0.01 --lower-bound ${MMLU_LOWER_BOUND:-0.38} --disable-tqdm"
 TP=${TP:-1} PP=${PP:-1} EP=${EP:-1} ETP=${ETP:-1} MLM_MODEL_CKPT=${MLM_MODEL_SAVE} ${MMLU_EXE} ${MLM_MODEL_CFG}
 
+# Export quantized checkpoint to HF format (PP=all GPUs)
+TOTAL_GPUS=$(python3 -c "import torch; print(torch.cuda.device_count())" 2>/dev/null || echo ${NUM_GPUS:-1})
+echo "=== Exporting ${MLM_MODEL_CFG} ${QUANT_CFG} (PP=${TOTAL_GPUS}) ==="
+export MLM_EXTRA_ARGS=
+TP=1 PP=${TOTAL_GPUS} EP=1 ETP=1 MLM_MODEL_CKPT=${MLM_MODEL_SAVE} ${EXPORT_EXE} ${MLM_MODEL_CFG}
+ls ${EXPORT_DIR}
+cat ${EXPORT_DIR}/hf_quant_config.json
+
 ###################################################################################################
 
 # This function handles the exit status (fails the CI).
diff --git a/tools/launcher/common/tensorrt_llm/eval.sh b/tools/launcher/common/tensorrt_llm/eval.sh
new file mode 100644
index 00000000000..3e3f2d1b768
--- /dev/null
+++ b/tools/launcher/common/tensorrt_llm/eval.sh
@@ -0,0 +1,59 @@
+#!/bin/bash
+
+# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+SCRIPT_DIR="$(dirname "$(readlink -f "$0")")"
+
+###################################################################################################
+
+if [[ -z ${HF_MODEL_CKPT} ]]; then
+    export HF_MODEL_CKPT=/scratchspace/export
+fi
+
+if [[ -z ${TP} ]]; then
+    TP=4
+fi
+
+if [[ -z ${EP} ]]; then
+    EP=4
+fi
+
+if [[ -z ${EXTRA_LLM_API_OPTIONS} ]]; then
+    EXTRA_LLM_API_OPTIONS=common/tensorrt_llm/extra_llm_api_options.yaml
+fi
+
+
+TARGET_FILENAME="config.json"
+
+
+# Find all files matching the target filename, print their paths null-terminated
+find "${HF_MODEL_CKPT}" -type f -name "$TARGET_FILENAME" -print0 | while IFS= read -r -d '' filepath; do
+    # Extract the directory path from the full file path
+    dir_path=$(dirname "$filepath")
+    
+    echo "Processing model: $dir_path"
+    # Place your commands here to run within or on the $dir_path
+    # Example: cd "$dir_path" && some_command
+
+    trtllm-llmapi-launch trtllm-eval \
+        --model ${dir_path} \
+        --disable_kv_cache_reuse \
+        --tp_size ${TP} \
+        --ep_size ${EP} \
+        --trust_remote_code \
+        --extra_llm_api_options ${EXTRA_LLM_API_OPTIONS} \
+        mmlu
+done
diff --git a/tools/launcher/common/tensorrt_llm/extra_llm_api_options.yaml b/tools/launcher/common/tensorrt_llm/extra_llm_api_options.yaml
new file mode 100644
index 00000000000..f5f43140861
--- /dev/null
+++ b/tools/launcher/common/tensorrt_llm/extra_llm_api_options.yaml
@@ -0,0 +1,52 @@
+context_parallel_size: 1
+
+  # backend: _autodeploy
+  # reasoning_parser: nano-v3
+  # tool_parser: qwen3_coder
+  # 
+  # runtime: trtllm
+  # compile_backend: torch-cudagraph
+  # max_batch_size: 64
+  # max_seq_len: 16384
+  # enable_chunked_prefill: true
+  # attn_backend: flashinfer
+  # model_factory: AutoModelForCausalLM
+  # skip_loading_weights: false
+  # free_mem_ratio: 0.65
+  # cuda_graph_batch_sizes: [1, 2, 4, 8, 16, 24, 32, 64, 128, 256, 320, 384]
+  # kv_cache_config:
+  #   # disable kv_cache reuse since not supported for hybrid/ssm models
+  #   enable_block_reuse: false
+  # transforms:
+  #   detect_sharding:
+  #     sharding_dims: ['ep', 'bmm']
+  #     allreduce_strategy: 'AUTO'
+  #     manual_config:
+  #       head_dim: 128
+  #       tp_plan:
+  #         # mamba SSM layer
+  #         "in_proj": "mamba"
+  #         "out_proj": "rowwise"
+  #         # attention layer
+  #         "q_proj": "colwise"
+  #         "k_proj": "colwise"
+  #         "v_proj": "colwise"
+  #         "o_proj": "rowwise"
+  #         # NOTE: consider not sharding shared experts and/or
+  #         # latent projections at all, keeping them replicated.
+  #         # To do so, comment out the corresponding entries.
+  #         # moe layer: SHARED experts
+  #         "up_proj": "colwise"
+  #         "down_proj": "rowwise"
+  #         # MoLE: latent projections: simple shard
+  #         "fc1_latent_proj": "gather"
+  #         "fc2_latent_proj": "gather"
+  #   multi_stream_moe:
+  #     stage: compile
+  #     enabled: true
+  #   insert_cached_ssm_attention:
+  #       cache_config:
+  #         mamba_dtype: float32
+  #   fuse_mamba_a_log:
+  #     stage: post_load_fusion
+  #     enabled: true
diff --git a/tools/launcher/examples/Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml b/tools/launcher/examples/Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml
index 7ba0566f254..0eeca6531c9 100644
--- a/tools/launcher/examples/Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml
+++ b/tools/launcher/examples/Qwen/Qwen3-30B-A3B/megatron_lm_ptq.yaml
@@ -21,7 +21,6 @@ pipeline:
       pp: 1
       ep: 8
       etp: 1
-      extra_args: ""
       calib_dataset: abisee/cnn_dailymail
       calib_size: 32
       mmlu_dataset: cais/mmlu
@@ -42,7 +41,6 @@ pipeline:
       pp: 1
       ep: 8
       etp: 1
-      extra_args: ""
       calib_dataset: abisee/cnn_dailymail
       calib_size: 32
       mmlu_dataset: cais/mmlu
diff --git a/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml b/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml
index 08d4dfccf2b..33b9da18e66 100644
--- a/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml
+++ b/tools/launcher/examples/Qwen/Qwen3-8B/megatron_lm_ptq.yaml
@@ -1,6 +1,9 @@
 # Qwen3-8B PTQ quantization (8 GPUs, for Slurm clusters).
 #
-# 2-step pipeline: NVFP4 then FP8, each followed by MMLU evaluation.
+# 3-step pipeline:
+#   task_0: NVFP4 quantize → MMLU → export
+#   task_1: FP8 quantize → MMLU → export
+#   task_2: TRT-LLM eval MMLU on all exported checkpoints
 #
 # Uses MegatronLMQuantizeTask with typed config — see common/megatron_lm/quantize/task.py
 # for all available fields.
@@ -49,3 +52,16 @@ pipeline:
       nodes: 1
       ntasks_per_node: 8
       gpus_per_node: 8
+
+  # Step 3: TRT-LLM eval MMLU on all exported checkpoints
+  task_2:
+    script: common/tensorrt_llm/eval.sh
+    environment:
+      - HF_MODEL_CKPT: /scratchspace/export
+      - TP: "8"
+      - EP: "1"
+    slurm_config:
+      _factory_: "slurm_factory"
+      nodes: 1
+      ntasks_per_node: 1
+      gpus_per_node: 8
diff --git a/uv.lock b/uv.lock
index 0f4710f92b4..cfa742f1081 100644
--- a/uv.lock
+++ b/uv.lock
@@ -32,9 +32,6 @@ resolution-markers = [
     "python_full_version < '3.11' and platform_machine == 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'",
 ]
 
-[manifest]
-overrides = [{ name = "torch", marker = "sys_platform == 'never'" }]
-
 [[package]]
 name = "accelerate"
 version = "1.13.0"
@@ -47,7 +44,7 @@ dependencies = [
     { name = "psutil" },
     { name = "pyyaml" },
     { name = "safetensors" },
-    { name = "torch", marker = "sys_platform == 'never'" },
+    { name = "torch" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/ca/14/787e5498cd062640f0f3d92ef4ae4063174f76f9afd29d13fc52a319daae/accelerate-1.13.0.tar.gz", hash = "sha256:d631b4e0f5b3de4aff2d7e9e6857d164810dfc3237d54d017f075122d057b236", size = 402835, upload-time = "2026-03-04T19:34:12.359Z" }
 wheels = [
@@ -573,6 +570,21 @@ dependencies = [
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/54/27/01d9078a77b9e31b79b9716e66ca4db74f4744c5232bcb3e8769395c4280/cppimport-22.8.2.tar.gz", hash = "sha256:bbb4957102db41bc99ad72c233bce92f9d1fd91be352fc07878c4361033a401f", size = 26635, upload-time = "2022-08-02T16:50:36.872Z" }
 
+[[package]]
+name = "cuda-bindings"
+version = "12.9.4"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "cuda-pathfinder", marker = "platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/7a/d8/b546104b8da3f562c1ff8ab36d130c8fe1dd6a045ced80b4f6ad74f7d4e1/cuda_bindings-12.9.4-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:4d3c842c2a4303b2a580fe955018e31aea30278be19795ae05226235268032e5", size = 12148218, upload-time = "2025-10-21T14:51:28.855Z" },
+    { url = "https://files.pythonhosted.org/packages/45/e7/b47792cc2d01c7e1d37c32402182524774dadd2d26339bd224e0e913832e/cuda_bindings-12.9.4-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:c912a3d9e6b6651853eed8eed96d6800d69c08e94052c292fec3f282c5a817c9", size = 12210593, upload-time = "2025-10-21T14:51:36.574Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/c1/dabe88f52c3e3760d861401bb994df08f672ec893b8f7592dc91626adcf3/cuda_bindings-12.9.4-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:fda147a344e8eaeca0c6ff113d2851ffca8f7dfc0a6c932374ee5c47caa649c8", size = 12151019, upload-time = "2025-10-21T14:51:43.167Z" },
+    { url = "https://files.pythonhosted.org/packages/63/56/e465c31dc9111be3441a9ba7df1941fe98f4aa6e71e8788a3fb4534ce24d/cuda_bindings-12.9.4-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:32bdc5a76906be4c61eb98f546a6786c5773a881f3b166486449b5d141e4a39f", size = 11906628, upload-time = "2025-10-21T14:51:49.905Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/84/1e6be415e37478070aeeee5884c2022713c1ecc735e6d82d744de0252eee/cuda_bindings-12.9.4-cp313-cp313t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:56e0043c457a99ac473ddc926fe0dc4046694d99caef633e92601ab52cbe17eb", size = 11925991, upload-time = "2025-10-21T14:51:56.535Z" },
+]
+
 [[package]]
 name = "cuda-pathfinder"
 version = "1.5.3"
@@ -637,18 +649,18 @@ name = "deepspeed"
 version = "0.18.9"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "einops", marker = "sys_platform != 'win32'" },
-    { name = "hjson", marker = "sys_platform != 'win32'" },
-    { name = "msgpack", marker = "sys_platform != 'win32'" },
-    { name = "ninja", marker = "sys_platform != 'win32'" },
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11' and sys_platform != 'win32'" },
-    { name = "numpy", version = "2.4.4", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11' and sys_platform != 'win32'" },
-    { name = "packaging", marker = "sys_platform != 'win32'" },
-    { name = "psutil", marker = "sys_platform != 'win32'" },
-    { name = "py-cpuinfo", marker = "sys_platform != 'win32'" },
-    { name = "pydantic", marker = "sys_platform != 'win32'" },
-    { name = "torch", marker = "sys_platform == 'never'" },
-    { name = "tqdm", marker = "sys_platform != 'win32'" },
+    { name = "einops", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "hjson", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "msgpack", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "ninja", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version < '3.11' and platform_machine != 's390x' and sys_platform == 'darwin') or (python_full_version < '3.11' and sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "numpy", version = "2.4.4", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.11' and platform_machine != 's390x' and sys_platform == 'darwin') or (python_full_version >= '3.11' and sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "packaging", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "psutil", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "py-cpuinfo", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "pydantic", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "torch", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
+    { name = "tqdm", marker = "(platform_machine != 's390x' and sys_platform == 'darwin') or (sys_platform != 'darwin' and sys_platform != 'win32')" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/24/61/5ea1c63b139fe7530b196b68ce0bdffa9cde79882e527dcecae58bd6c770/deepspeed-0.18.9.tar.gz", hash = "sha256:ee4818dcf342794f74f429a0aeebef90291ec808fa82609c5140c23e665c4011", size = 1663466, upload-time = "2026-03-30T16:43:16.566Z" }
 
@@ -1520,7 +1532,11 @@ name = "networkx"
 version = "3.4.2"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
+    "python_full_version < '3.11' and platform_machine == 'aarch64' and sys_platform == 'win32'",
     "(python_full_version < '3.11' and platform_machine != 's390x' and sys_platform == 'darwin') or (python_full_version < '3.11' and platform_machine == 'aarch64' and sys_platform != 'darwin' and sys_platform != 'win32')",
+    "python_full_version < '3.11' and platform_machine == 's390x' and sys_platform == 'darwin'",
+    "python_full_version < '3.11' and platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform == 'win32'",
+    "python_full_version < '3.11' and platform_machine == 's390x' and sys_platform == 'win32'",
     "python_full_version < '3.11' and platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'",
     "python_full_version < '3.11' and platform_machine == 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'",
 ]
@@ -1534,15 +1550,27 @@ name = "networkx"
 version = "3.6.1"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
+    "python_full_version >= '3.13' and platform_machine == 'aarch64' and sys_platform == 'win32'",
+    "python_full_version == '3.12.*' and platform_machine == 'aarch64' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and platform_machine == 'aarch64' and sys_platform == 'win32'",
     "(python_full_version >= '3.13' and platform_machine != 's390x' and sys_platform == 'darwin') or (python_full_version >= '3.13' and platform_machine == 'aarch64' and sys_platform != 'darwin' and sys_platform != 'win32')",
+    "python_full_version >= '3.13' and platform_machine == 's390x' and sys_platform == 'darwin'",
     "(python_full_version == '3.12.*' and platform_machine != 's390x' and sys_platform == 'darwin') or (python_full_version == '3.12.*' and platform_machine == 'aarch64' and sys_platform != 'darwin' and sys_platform != 'win32')",
+    "python_full_version == '3.12.*' and platform_machine == 's390x' and sys_platform == 'darwin'",
     "(python_full_version == '3.11.*' and platform_machine != 's390x' and sys_platform == 'darwin') or (python_full_version == '3.11.*' and platform_machine == 'aarch64' and sys_platform != 'darwin' and sys_platform != 'win32')",
+    "python_full_version == '3.11.*' and platform_machine == 's390x' and sys_platform == 'darwin'",
     "python_full_version >= '3.13' and platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'",
     "python_full_version >= '3.13' and platform_machine == 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'",
     "python_full_version == '3.12.*' and platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'",
     "python_full_version == '3.12.*' and platform_machine == 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'",
     "python_full_version == '3.11.*' and platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'",
     "python_full_version == '3.11.*' and platform_machine == 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'",
+    "python_full_version >= '3.13' and platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform == 'win32'",
+    "python_full_version >= '3.13' and platform_machine == 's390x' and sys_platform == 'win32'",
+    "python_full_version == '3.12.*' and platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform == 'win32'",
+    "python_full_version == '3.12.*' and platform_machine == 's390x' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and platform_machine == 's390x' and sys_platform == 'win32'",
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/6a/51/63fe664f3908c97be9d2e4f1158eb633317598cfa6e1fc14af5383f17512/networkx-3.6.1.tar.gz", hash = "sha256:26b7c357accc0c8cde558ad486283728b65b6a95d85ee1cd66bafab4c8168509", size = 2517025, upload-time = "2025-12-08T17:02:39.908Z" }
 wheels = [
@@ -1751,6 +1779,108 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/04/74/f4c001f4714c3ad9ce037e18cf2b9c64871a84951eaa0baf683a9ca9301c/numpy-2.4.4-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:f2cf083b324a467e1ab358c105f6cad5ea950f50524668a80c486ff1db24e119", size = 12509075, upload-time = "2026-03-29T13:21:57.644Z" },
 ]
 
+[[package]]
+name = "nvidia-cublas-cu12"
+version = "12.8.4.1"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/dc/61/e24b560ab2e2eaeb3c839129175fb330dfcfc29e5203196e5541a4c44682/nvidia_cublas_cu12-12.8.4.1-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:8ac4e771d5a348c551b2a426eda6193c19aa630236b418086020df5ba9667142", size = 594346921, upload-time = "2025-03-07T01:44:31.254Z" },
+]
+
+[[package]]
+name = "nvidia-cuda-cupti-cu12"
+version = "12.8.90"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f8/02/2adcaa145158bf1a8295d83591d22e4103dbfd821bcaf6f3f53151ca4ffa/nvidia_cuda_cupti_cu12-12.8.90-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:ea0cb07ebda26bb9b29ba82cda34849e73c166c18162d3913575b0c9db9a6182", size = 10248621, upload-time = "2025-03-07T01:40:21.213Z" },
+]
+
+[[package]]
+name = "nvidia-cuda-nvrtc-cu12"
+version = "12.8.93"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/05/6b/32f747947df2da6994e999492ab306a903659555dddc0fbdeb9d71f75e52/nvidia_cuda_nvrtc_cu12-12.8.93-py3-none-manylinux2010_x86_64.manylinux_2_12_x86_64.whl", hash = "sha256:a7756528852ef889772a84c6cd89d41dfa74667e24cca16bb31f8f061e3e9994", size = 88040029, upload-time = "2025-03-07T01:42:13.562Z" },
+]
+
+[[package]]
+name = "nvidia-cuda-runtime-cu12"
+version = "12.8.90"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/0d/9b/a997b638fcd068ad6e4d53b8551a7d30fe8b404d6f1804abf1df69838932/nvidia_cuda_runtime_cu12-12.8.90-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:adade8dcbd0edf427b7204d480d6066d33902cab2a4707dcfc48a2d0fd44ab90", size = 954765, upload-time = "2025-03-07T01:40:01.615Z" },
+]
+
+[[package]]
+name = "nvidia-cudnn-cu12"
+version = "9.10.2.21"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "nvidia-cublas-cu12", marker = "platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ba/51/e123d997aa098c61d029f76663dedbfb9bc8dcf8c60cbd6adbe42f76d049/nvidia_cudnn_cu12-9.10.2.21-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:949452be657fa16687d0930933f032835951ef0892b37d2d53824d1a84dc97a8", size = 706758467, upload-time = "2025-06-06T21:54:08.597Z" },
+]
+
+[[package]]
+name = "nvidia-cufft-cu12"
+version = "11.3.3.83"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "nvidia-nvjitlink-cu12", marker = "platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/1f/13/ee4e00f30e676b66ae65b4f08cb5bcbb8392c03f54f2d5413ea99a5d1c80/nvidia_cufft_cu12-11.3.3.83-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:4d2dd21ec0b88cf61b62e6b43564355e5222e4a3fb394cac0db101f2dd0d4f74", size = 193118695, upload-time = "2025-03-07T01:45:27.821Z" },
+]
+
+[[package]]
+name = "nvidia-cufile-cu12"
+version = "1.13.1.3"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/bb/fe/1bcba1dfbfb8d01be8d93f07bfc502c93fa23afa6fd5ab3fc7c1df71038a/nvidia_cufile_cu12-1.13.1.3-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:1d069003be650e131b21c932ec3d8969c1715379251f8d23a1860554b1cb24fc", size = 1197834, upload-time = "2025-03-07T01:45:50.723Z" },
+]
+
+[[package]]
+name = "nvidia-curand-cu12"
+version = "10.3.9.90"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/fb/aa/6584b56dc84ebe9cf93226a5cde4d99080c8e90ab40f0c27bda7a0f29aa1/nvidia_curand_cu12-10.3.9.90-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:b32331d4f4df5d6eefa0554c565b626c7216f87a06a4f56fab27c3b68a830ec9", size = 63619976, upload-time = "2025-03-07T01:46:23.323Z" },
+]
+
+[[package]]
+name = "nvidia-cusolver-cu12"
+version = "11.7.3.90"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "nvidia-cublas-cu12", marker = "platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'" },
+    { name = "nvidia-cusparse-cu12", marker = "platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'" },
+    { name = "nvidia-nvjitlink-cu12", marker = "platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/85/48/9a13d2975803e8cf2777d5ed57b87a0b6ca2cc795f9a4f59796a910bfb80/nvidia_cusolver_cu12-11.7.3.90-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:4376c11ad263152bd50ea295c05370360776f8c3427b30991df774f9fb26c450", size = 267506905, upload-time = "2025-03-07T01:47:16.273Z" },
+]
+
+[[package]]
+name = "nvidia-cusparse-cu12"
+version = "12.5.8.93"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "nvidia-nvjitlink-cu12", marker = "platform_machine != 'aarch64' and platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'win32'" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c2/f5/e1854cb2f2bcd4280c44736c93550cc300ff4b8c95ebe370d0aa7d2b473d/nvidia_cusparse_cu12-12.5.8.93-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:1ec05d76bbbd8b61b06a80e1eaf8cf4959c3d4ce8e711b65ebd0443bb0ebb13b", size = 288216466, upload-time = "2025-03-07T01:48:13.779Z" },
+]
+
+[[package]]
+name = "nvidia-cusparselt-cu12"
+version = "0.7.1"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/56/79/12978b96bd44274fe38b5dde5cfb660b1d114f70a65ef962bcbbed99b549/nvidia_cusparselt_cu12-0.7.1-py3-none-manylinux2014_x86_64.whl", hash = "sha256:f1bb701d6b930d5a7cea44c19ceb973311500847f81b634d802b7b539dc55623", size = 287193691, upload-time = "2025-02-26T00:15:44.104Z" },
+]
+
 [[package]]
 name = "nvidia-ml-py"
 version = "13.595.45"
@@ -1779,7 +1909,7 @@ dependencies = [
     { name = "scipy", version = "1.15.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
     { name = "scipy", version = "1.17.1", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
     { name = "setuptools" },
-    { name = "torch", marker = "sys_platform == 'never'" },
+    { name = "torch" },
     { name = "tqdm" },
 ]
 
@@ -2026,6 +2156,38 @@ requires-dist = [
 ]
 provides-extras = ["onnx", "hf", "puzzletron", "dev-lint", "dev-docs", "dev-test", "all", "dev"]
 
+[[package]]
+name = "nvidia-nccl-cu12"
+version = "2.27.5"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/6e/89/f7a07dc961b60645dbbf42e80f2bc85ade7feb9a491b11a1e973aa00071f/nvidia_nccl_cu12-2.27.5-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:ad730cf15cb5d25fe849c6e6ca9eb5b76db16a80f13f425ac68d8e2e55624457", size = 322348229, upload-time = "2025-06-26T04:11:28.385Z" },
+]
+
+[[package]]
+name = "nvidia-nvjitlink-cu12"
+version = "12.8.93"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f6/74/86a07f1d0f42998ca31312f998bd3b9a7eff7f52378f4f270c8679c77fb9/nvidia_nvjitlink_cu12-12.8.93-py3-none-manylinux2010_x86_64.manylinux_2_12_x86_64.whl", hash = "sha256:81ff63371a7ebd6e6451970684f916be2eab07321b73c9d244dc2b4da7f73b88", size = 39254836, upload-time = "2025-03-07T01:49:55.661Z" },
+]
+
+[[package]]
+name = "nvidia-nvshmem-cu12"
+version = "3.4.5"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b5/09/6ea3ea725f82e1e76684f0708bbedd871fc96da89945adeba65c3835a64c/nvidia_nvshmem_cu12-3.4.5-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:042f2500f24c021db8a06c5eec2539027d57460e1c1a762055a6554f72c369bd", size = 139103095, upload-time = "2025-09-06T00:32:31.266Z" },
+]
+
+[[package]]
+name = "nvidia-nvtx-cu12"
+version = "12.8.90"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/a2/eb/86626c1bbc2edb86323022371c39aa48df6fd8b0a1647bc274577f72e90b/nvidia_nvtx_cu12-12.8.90-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:5b17e2001cc0d751a5bc2c6ec6d26ad95913324a4adb86788c944f8ce9ba441f", size = 89954, upload-time = "2025-03-07T01:42:44.131Z" },
+]
+
 [[package]]
 name = "omegaconf"
 version = "2.3.0"
@@ -2442,7 +2604,7 @@ dependencies = [
     { name = "psutil" },
     { name = "pyyaml" },
     { name = "safetensors" },
-    { name = "torch", marker = "sys_platform == 'never'" },
+    { name = "torch" },
     { name = "tqdm" },
     { name = "transformers" },
 ]
@@ -3782,7 +3944,7 @@ dependencies = [
     { name = "huggingface-hub" },
     { name = "pyyaml" },
     { name = "safetensors" },
-    { name = "torch", marker = "sys_platform == 'never'" },
+    { name = "torch" },
     { name = "torchvision" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/7b/1e/e924b3b2326a856aaf68586f9c52a5fc81ef45715eca408393b68c597e0e/timm-1.0.26.tar.gz", hash = "sha256:f66f082f2f381cf68431c22714c8b70f723837fa2a185b155961eab90f2d5b10", size = 2419859, upload-time = "2026-03-23T18:12:10.272Z" }
@@ -3870,15 +4032,63 @@ name = "torch"
 version = "2.10.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
+    { name = "cuda-bindings", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
     { name = "filelock" },
     { name = "fsspec" },
     { name = "jinja2" },
     { name = "networkx", version = "3.4.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
     { name = "networkx", version = "3.6.1", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "nvidia-cublas-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-cuda-cupti-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-cuda-nvrtc-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-cuda-runtime-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-cudnn-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-cufft-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-cufile-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-curand-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-cusolver-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-cusparse-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-cusparselt-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-nccl-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-nvjitlink-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-nvshmem-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "nvidia-nvtx-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
     { name = "setuptools", marker = "python_full_version >= '3.12'" },
     { name = "sympy" },
+    { name = "triton", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
     { name = "typing-extensions" },
 ]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/5b/30/bfebdd8ec77db9a79775121789992d6b3b75ee5494971294d7b4b7c999bc/torch-2.10.0-2-cp310-none-macosx_11_0_arm64.whl", hash = "sha256:2b980edd8d7c0a68c4e951ee1856334a43193f98730d97408fbd148c1a933313", size = 79411457, upload-time = "2026-02-10T21:44:59.189Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/8b/4b61d6e13f7108f36910df9ab4b58fd389cc2520d54d81b88660804aad99/torch-2.10.0-2-cp311-none-macosx_11_0_arm64.whl", hash = "sha256:418997cb02d0a0f1497cf6a09f63166f9f5df9f3e16c8a716ab76a72127c714f", size = 79423467, upload-time = "2026-02-10T21:44:48.711Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/54/a2ba279afcca44bbd320d4e73675b282fcee3d81400ea1b53934efca6462/torch-2.10.0-2-cp312-none-macosx_11_0_arm64.whl", hash = "sha256:13ec4add8c3faaed8d13e0574f5cd4a323c11655546f91fbe6afa77b57423574", size = 79498202, upload-time = "2026-02-10T21:44:52.603Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/23/2c9fe0c9c27f7f6cb865abcea8a4568f29f00acaeadfc6a37f6801f84cb4/torch-2.10.0-2-cp313-none-macosx_11_0_arm64.whl", hash = "sha256:e521c9f030a3774ed770a9c011751fb47c4d12029a3d6522116e48431f2ff89e", size = 79498254, upload-time = "2026-02-10T21:44:44.095Z" },
+    { url = "https://files.pythonhosted.org/packages/16/ee/efbd56687be60ef9af0c9c0ebe106964c07400eade5b0af8902a1d8cd58c/torch-2.10.0-3-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:a1ff626b884f8c4e897c4c33782bdacdff842a165fee79817b1dd549fdda1321", size = 915510070, upload-time = "2026-03-11T14:16:39.386Z" },
+    { url = "https://files.pythonhosted.org/packages/36/ab/7b562f1808d3f65414cd80a4f7d4bb00979d9355616c034c171249e1a303/torch-2.10.0-3-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:ac5bdcbb074384c66fa160c15b1ead77839e3fe7ed117d667249afce0acabfac", size = 915518691, upload-time = "2026-03-11T14:15:43.147Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/7a/abada41517ce0011775f0f4eacc79659bc9bc6c361e6bfe6f7052a6b9363/torch-2.10.0-3-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:98c01b8bb5e3240426dcde1446eed6f40c778091c8544767ef1168fc663a05a6", size = 915622781, upload-time = "2026-03-11T14:17:11.354Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/c6/4dfe238342ffdcec5aef1c96c457548762d33c40b45a1ab7033bb26d2ff2/torch-2.10.0-3-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:80b1b5bfe38eb0e9f5ff09f206dcac0a87aadd084230d4a36eea5ec5232c115b", size = 915627275, upload-time = "2026-03-11T14:16:11.325Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/f0/72bf18847f58f877a6a8acf60614b14935e2f156d942483af1ffc081aea0/torch-2.10.0-3-cp313-cp313t-manylinux_2_28_x86_64.whl", hash = "sha256:46b3574d93a2a8134b3f5475cfb98e2eb46771794c57015f6ad1fb795ec25e49", size = 915523474, upload-time = "2026-03-11T14:17:44.422Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/1a/c61f36cfd446170ec27b3a4984f072fd06dab6b5d7ce27e11adb35d6c838/torch-2.10.0-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:5276fa790a666ee8becaffff8acb711922252521b28fbce5db7db5cf9cb2026d", size = 145992962, upload-time = "2026-01-21T16:24:14.04Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/60/6662535354191e2d1555296045b63e4279e5a9dbad49acf55a5d38655a39/torch-2.10.0-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:aaf663927bcd490ae971469a624c322202a2a1e68936eb952535ca4cd3b90444", size = 915599237, upload-time = "2026-01-21T16:23:25.497Z" },
+    { url = "https://files.pythonhosted.org/packages/40/b8/66bbe96f0d79be2b5c697b2e0b187ed792a15c6c4b8904613454651db848/torch-2.10.0-cp310-cp310-win_amd64.whl", hash = "sha256:a4be6a2a190b32ff5c8002a0977a25ea60e64f7ba46b1be37093c141d9c49aeb", size = 113720931, upload-time = "2026-01-21T16:24:23.743Z" },
+    { url = "https://files.pythonhosted.org/packages/76/bb/d820f90e69cda6c8169b32a0c6a3ab7b17bf7990b8f2c680077c24a3c14c/torch-2.10.0-cp310-none-macosx_11_0_arm64.whl", hash = "sha256:35e407430795c8d3edb07a1d711c41cc1f9eaddc8b2f1cc0a165a6767a8fb73d", size = 79411450, upload-time = "2026-01-21T16:25:30.692Z" },
+    { url = "https://files.pythonhosted.org/packages/78/89/f5554b13ebd71e05c0b002f95148033e730d3f7067f67423026cc9c69410/torch-2.10.0-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:3282d9febd1e4e476630a099692b44fdc214ee9bf8ee5377732d9d9dfe5712e4", size = 145992610, upload-time = "2026-01-21T16:25:26.327Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/30/a3a2120621bf9c17779b169fc17e3dc29b230c29d0f8222f499f5e159aa8/torch-2.10.0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:a2f9edd8dbc99f62bc4dfb78af7bf89499bca3d753423ac1b4e06592e467b763", size = 915607863, upload-time = "2026-01-21T16:25:06.696Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/3d/c87b33c5f260a2a8ad68da7147e105f05868c281c63d65ed85aa4da98c66/torch-2.10.0-cp311-cp311-win_amd64.whl", hash = "sha256:29b7009dba4b7a1c960260fc8ac85022c784250af43af9fb0ebafc9883782ebd", size = 113723116, upload-time = "2026-01-21T16:25:21.916Z" },
+    { url = "https://files.pythonhosted.org/packages/61/d8/15b9d9d3a6b0c01b883787bd056acbe5cc321090d4b216d3ea89a8fcfdf3/torch-2.10.0-cp311-none-macosx_11_0_arm64.whl", hash = "sha256:b7bd80f3477b830dd166c707c5b0b82a898e7b16f59a7d9d42778dd058272e8b", size = 79423461, upload-time = "2026-01-21T16:24:50.266Z" },
+    { url = "https://files.pythonhosted.org/packages/cc/af/758e242e9102e9988969b5e621d41f36b8f258bb4a099109b7a4b4b50ea4/torch-2.10.0-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:5fd4117d89ffd47e3dcc71e71a22efac24828ad781c7e46aaaf56bf7f2796acf", size = 145996088, upload-time = "2026-01-21T16:24:44.171Z" },
+    { url = "https://files.pythonhosted.org/packages/23/8e/3c74db5e53bff7ed9e34c8123e6a8bfef718b2450c35eefab85bb4a7e270/torch-2.10.0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:787124e7db3b379d4f1ed54dd12ae7c741c16a4d29b49c0226a89bea50923ffb", size = 915711952, upload-time = "2026-01-21T16:23:53.503Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/01/624c4324ca01f66ae4c7cd1b74eb16fb52596dce66dbe51eff95ef9e7a4c/torch-2.10.0-cp312-cp312-win_amd64.whl", hash = "sha256:2c66c61f44c5f903046cc696d088e21062644cbe541c7f1c4eaae88b2ad23547", size = 113757972, upload-time = "2026-01-21T16:24:39.516Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/5c/dee910b87c4d5c0fcb41b50839ae04df87c1cfc663cf1b5fca7ea565eeaa/torch-2.10.0-cp312-none-macosx_11_0_arm64.whl", hash = "sha256:6d3707a61863d1c4d6ebba7be4ca320f42b869ee657e9b2c21c736bf17000294", size = 79498198, upload-time = "2026-01-21T16:24:34.704Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/6f/f2e91e34e3fcba2e3fc8d8f74e7d6c22e74e480bbd1db7bc8900fdf3e95c/torch-2.10.0-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:5c4d217b14741e40776dd7074d9006fd28b8a97ef5654db959d8635b2fe5f29b", size = 146004247, upload-time = "2026-01-21T16:24:29.335Z" },
+    { url = "https://files.pythonhosted.org/packages/98/fb/5160261aeb5e1ee12ee95fe599d0541f7c976c3701d607d8fc29e623229f/torch-2.10.0-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:6b71486353fce0f9714ca0c9ef1c850a2ae766b409808acd58e9678a3edb7738", size = 915716445, upload-time = "2026-01-21T16:22:45.353Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/16/502fb1b41e6d868e8deb5b0e3ae926bbb36dab8ceb0d1b769b266ad7b0c3/torch-2.10.0-cp313-cp313-win_amd64.whl", hash = "sha256:c2ee399c644dc92ef7bc0d4f7e74b5360c37cdbe7c5ba11318dda49ffac2bc57", size = 113757050, upload-time = "2026-01-21T16:24:19.204Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/0b/39929b148f4824bc3ad6f9f72a29d4ad865bcf7ebfc2fa67584773e083d2/torch-2.10.0-cp313-cp313t-macosx_14_0_arm64.whl", hash = "sha256:3202429f58309b9fa96a614885eace4b7995729f44beb54d3e4a47773649d382", size = 79851305, upload-time = "2026-01-21T16:24:09.209Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/14/21fbce63bc452381ba5f74a2c0a959fdf5ad5803ccc0c654e752e0dbe91a/torch-2.10.0-cp313-cp313t-manylinux_2_28_aarch64.whl", hash = "sha256:aae1b29cd68e50a9397f5ee897b9c24742e9e306f88a807a27d617f07adb3bd8", size = 146005472, upload-time = "2026-01-21T16:22:29.022Z" },
+    { url = "https://files.pythonhosted.org/packages/54/fd/b207d1c525cb570ef47f3e9f836b154685011fce11a2f444ba8a4084d042/torch-2.10.0-cp313-cp313t-manylinux_2_28_x86_64.whl", hash = "sha256:6021db85958db2f07ec94e1bc77212721ba4920c12a18dc552d2ae36a3eb163f", size = 915612644, upload-time = "2026-01-21T16:21:47.019Z" },
+    { url = "https://files.pythonhosted.org/packages/36/53/0197f868c75f1050b199fe58f9bf3bf3aecac9b4e85cc9c964383d745403/torch-2.10.0-cp313-cp313t-win_amd64.whl", hash = "sha256:ff43db38af76fda183156153983c9a096fc4c78d0cd1e07b14a2314c7f01c2c8", size = 113997015, upload-time = "2026-01-21T16:23:00.767Z" },
+    { url = "https://files.pythonhosted.org/packages/0e/13/e76b4d9c160e89fff48bf16b449ea324bda84745d2ab30294c37c2434c0d/torch-2.10.0-cp313-none-macosx_11_0_arm64.whl", hash = "sha256:cdf2a523d699b70d613243211ecaac14fe9c5df8a0b0a9c02add60fb2a413e0f", size = 79498248, upload-time = "2026-01-21T16:23:09.315Z" },
+]
 
 [[package]]
 name = "torch-geometric"
@@ -3908,7 +4118,7 @@ source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
     { name = "numpy", version = "2.4.4", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "torch", marker = "sys_platform == 'never'" },
+    { name = "torch" },
     { name = "torchvision" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/6f/36/574c0c46e818533b78b3c09505211162918188325ab4165ef11a3f295755/torchprofile-0.0.4.tar.gz", hash = "sha256:96b6da17d752a06b02977e078aea95614893b31d4117dd5dcd081f30ce65611b", size = 4557, upload-time = "2021-06-22T04:58:03.592Z" }
@@ -3918,35 +4128,35 @@ wheels = [
 
 [[package]]
 name = "torchvision"
-version = "0.26.0"
+version = "0.25.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
     { name = "numpy", version = "2.4.4", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
     { name = "pillow" },
-    { name = "torch", marker = "sys_platform == 'never'" },
-]
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/74/b4/cdfee31e0402ea035135462cb0ab496e974d56fab6b4e7a1f0cbccb8cd28/torchvision-0.26.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:a06d4772a8e13e772906ed736cc53ec6639e5e60554f8e5fa6ca165aabebc464", size = 1863503, upload-time = "2026-03-23T18:13:01.384Z" },
-    { url = "https://files.pythonhosted.org/packages/e4/74/11fee109841e80ad14e5ca2d80bff6b10eb11b7838ff06f35bfeaa9f7251/torchvision-0.26.0-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:2adfbe438473236191ff077a4a9a0c767436879c89628aa97137e959b0c11a94", size = 7766423, upload-time = "2026-03-23T18:12:56.049Z" },
-    { url = "https://files.pythonhosted.org/packages/5e/00/24d8c7845c3f270153fb81395a5135b2778e2538e81d14c6aea5106c689c/torchvision-0.26.0-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:b6f9ad1ecc0eab52647298b379ee9426845f8903703e6127973f8f3d049a798b", size = 7518249, upload-time = "2026-03-23T18:12:51.743Z" },
-    { url = "https://files.pythonhosted.org/packages/d7/ed/e53cd7c0da7ae002e5e929c1796ebbe7ec0c700c29f7a0a6696497fb3d8b/torchvision-0.26.0-cp310-cp310-win_amd64.whl", hash = "sha256:f13f12b3791a266de2d599cb8162925261622a037d87fc03132848343cf68f75", size = 3669784, upload-time = "2026-03-23T18:12:49.949Z" },
-    { url = "https://files.pythonhosted.org/packages/b4/bd/d552a2521bade3295b2c6e7a4a0d1022261cab7ca7011f4e2a330dbb3caa/torchvision-0.26.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:55bd6ad4ae77be01ba67a410b05b51f53b0d0ee45f146eb6a0dfb9007e70ab3c", size = 1863499, upload-time = "2026-03-23T18:12:58.696Z" },
-    { url = "https://files.pythonhosted.org/packages/33/bf/21b899792b08cae7a298551c68398a79e333697479ed311b3b067aab4bdc/torchvision-0.26.0-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:1c55dc8affbcc0eb2060fbabbe996ae9e5839b24bb6419777f17848945a411b1", size = 7767527, upload-time = "2026-03-23T18:12:44.348Z" },
-    { url = "https://files.pythonhosted.org/packages/9a/45/57bbf9e216850d065e66dd31a50f57424b607f1d878ab8956e56a1f4e36b/torchvision-0.26.0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:fd10b5f994c210f4f6d6761cf686f82d748554adf486cb0979770c3252868c8f", size = 7519925, upload-time = "2026-03-23T18:12:53.283Z" },
-    { url = "https://files.pythonhosted.org/packages/10/58/ed8f7754299f3e91d6414b6dc09f62b3fa7c6e5d63dfe48d69ab81498a37/torchvision-0.26.0-cp311-cp311-win_amd64.whl", hash = "sha256:de6424b12887ad884f39a0ee446994ae3cd3b6a00a9cafe1bead85a031132af0", size = 3983834, upload-time = "2026-03-23T18:13:00.224Z" },
-    { url = "https://files.pythonhosted.org/packages/ae/e7/56b47cc3b132aea90ccce22bcb8975dec688b002150012acc842846039d0/torchvision-0.26.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:c409e1c3fdebec7a3834465086dbda8bf7680eff79abf7fd2f10c6b59520a7a4", size = 1863502, upload-time = "2026-03-23T18:12:57.326Z" },
-    { url = "https://files.pythonhosted.org/packages/f4/ec/5c31c92c08b65662fe9604a4067ae8232582805949f11ddc042cebe818ed/torchvision-0.26.0-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:406557718e62fdf10f5706e88d8a5ec000f872da913bf629aab9297622585547", size = 7767944, upload-time = "2026-03-23T18:12:42.805Z" },
-    { url = "https://files.pythonhosted.org/packages/f5/d8/cb6ccda1a1f35a6597645818641701207b3e8e13553e75fce5d86bac74b2/torchvision-0.26.0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:d61a5abb6b42a0c0c311996c2ac4b83a94418a97182c83b055a2a4ae985e05aa", size = 7522205, upload-time = "2026-03-23T18:12:54.654Z" },
-    { url = "https://files.pythonhosted.org/packages/1c/a9/c272623a0f735c35f0f6cd6dc74784d4f970e800cf063bb76687895a2ab9/torchvision-0.26.0-cp312-cp312-win_amd64.whl", hash = "sha256:7993c01648e7c61d191b018e84d38fe0825c8fcb2720cd0f37caf7ba14404aa1", size = 4255155, upload-time = "2026-03-23T18:12:32.652Z" },
-    { url = "https://files.pythonhosted.org/packages/da/80/0762f77f53605d10c9477be39bb47722cc8e383bbbc2531471ce0e396c07/torchvision-0.26.0-cp313-cp313-macosx_12_0_arm64.whl", hash = "sha256:5d63dd43162691258b1b3529b9041bac7d54caa37eae0925f997108268cbf7c4", size = 1860809, upload-time = "2026-03-23T18:12:47.629Z" },
-    { url = "https://files.pythonhosted.org/packages/e6/81/0b3e58d1478c660a5af4268713486b2df7203f35abd9195fea87348a5178/torchvision-0.26.0-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:a39c7a26538c41fda453f9a9692b5ff9b35a5437db1d94f3027f6f509c160eac", size = 7727494, upload-time = "2026-03-23T18:12:46.062Z" },
-    { url = "https://files.pythonhosted.org/packages/b6/dc/d9ab5d29115aa05e12e30f1397a3eeae1d88a511241dc3bce48dc4342675/torchvision-0.26.0-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:b7e6213620bbf97742e5f79832f9e9d769e6cf0f744c5b53dad80b76db633691", size = 7521747, upload-time = "2026-03-23T18:12:36.815Z" },
-    { url = "https://files.pythonhosted.org/packages/a9/1b/f1bc86a918c5f6feab1eeff11982e2060f4704332e96185463d27855bdf5/torchvision-0.26.0-cp313-cp313-win_amd64.whl", hash = "sha256:4280c35ec8cba1fcc8294fb87e136924708726864c379e4c54494797d86bc474", size = 4319880, upload-time = "2026-03-23T18:12:38.168Z" },
-    { url = "https://files.pythonhosted.org/packages/66/28/b4ad0a723ed95b003454caffcc41894b34bd8379df340848cae2c33871de/torchvision-0.26.0-cp313-cp313t-macosx_12_0_arm64.whl", hash = "sha256:358fc4726d0c08615b6d83b3149854f11efb2a564ed1acb6fce882e151412d23", size = 1951973, upload-time = "2026-03-23T18:12:48.781Z" },
-    { url = "https://files.pythonhosted.org/packages/71/e2/7a89096e6cf2f3336353b5338ba925e0addf9d8601920340e6bdf47e8eb3/torchvision-0.26.0-cp313-cp313t-manylinux_2_28_aarch64.whl", hash = "sha256:3daf9cc149cf3cdcbd4df9c59dae69ffca86c6823250442c3bbfd63fc2e26c61", size = 7728679, upload-time = "2026-03-23T18:12:26.196Z" },
-    { url = "https://files.pythonhosted.org/packages/69/1d/4e1eebc17d18ce080a11dcf3df3f8f717f0efdfa00983f06e8ba79259f61/torchvision-0.26.0-cp313-cp313t-manylinux_2_28_x86_64.whl", hash = "sha256:82c3965eca27e86a316e31e4c3e5a16d353e0bcbe0ef8efa2e66502c54493c4b", size = 7609138, upload-time = "2026-03-23T18:12:35.327Z" },
-    { url = "https://files.pythonhosted.org/packages/f3/a4/f1155e943ae5b32400d7000adc81c79bb0392b16ceb33bcf13e02e48cced/torchvision-0.26.0-cp313-cp313t-win_amd64.whl", hash = "sha256:ebc043cc5a4f0bf22e7680806dbba37ffb19e70f6953bbb44ed1a90aeb5c9bea", size = 4248202, upload-time = "2026-03-23T18:12:41.423Z" },
+    { name = "torch" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/50/ae/cbf727421eb73f1cf907fbe5788326a08f111b3f6b6ddca15426b53fec9a/torchvision-0.25.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:a95c47abb817d4e90ea1a8e57bd0d728e3e6b533b3495ae77d84d883c4d11f56", size = 1874919, upload-time = "2026-01-21T16:27:47.617Z" },
+    { url = "https://files.pythonhosted.org/packages/64/68/dc7a224f606d53ea09f9a85196a3921ec3a801b0b1d17e84c73392f0c029/torchvision-0.25.0-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:acc339aba4a858192998c2b91f635827e40d9c469d9cf1455bafdda6e4c28ea4", size = 2343220, upload-time = "2026-01-21T16:27:44.26Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/fa/8cce5ca7ffd4da95193232493703d20aa06303f37b119fd23a65df4f239a/torchvision-0.25.0-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:0d9a3f925a081dd2ebb0b791249b687c2ef2c2717d027946654607494b9b64b6", size = 8068106, upload-time = "2026-01-21T16:27:37.805Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/b9/a53bcf8f78f2cd89215e9ded70041765d50ef13bf301f9884ec6041a9421/torchvision-0.25.0-cp310-cp310-win_amd64.whl", hash = "sha256:b57430fbe9e9b697418a395041bb615124d9c007710a2712fda6e35fb310f264", size = 3697295, upload-time = "2026-01-21T16:27:36.574Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/be/c704bceaf11c4f6b19d64337a34a877fcdfe3bd68160a8c9ae9bea4a35a3/torchvision-0.25.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:db74a551946b75d19f9996c419a799ffdf6a223ecf17c656f90da011f1d75b20", size = 1874923, upload-time = "2026-01-21T16:27:46.574Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/e9/f143cd71232430de1f547ceab840f68c55e127d72558b1061a71d0b193cd/torchvision-0.25.0-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:f49964f96644dbac2506dffe1a0a7ec0f2bf8cf7a588c3319fed26e6329ffdf3", size = 2344808, upload-time = "2026-01-21T16:27:43.191Z" },
+    { url = "https://files.pythonhosted.org/packages/43/ae/ad5d6165797de234c9658752acb4fce65b78a6a18d82efdf8367c940d8da/torchvision-0.25.0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:153c0d2cbc34b7cf2da19d73450f24ba36d2b75ec9211b9962b5022fb9e4ecee", size = 8070752, upload-time = "2026-01-21T16:27:33.748Z" },
+    { url = "https://files.pythonhosted.org/packages/23/19/55b28aecdc7f38df57b8eb55eb0b14a62b470ed8efeb22cdc74224df1d6a/torchvision-0.25.0-cp311-cp311-win_amd64.whl", hash = "sha256:ea580ffd6094cc01914ad32f8c8118174f18974629af905cea08cb6d5d48c7b7", size = 4038722, upload-time = "2026-01-21T16:27:41.355Z" },
+    { url = "https://files.pythonhosted.org/packages/56/3a/6ea0d73f49a9bef38a1b3a92e8dd455cea58470985d25635beab93841748/torchvision-0.25.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:c2abe430c90b1d5e552680037d68da4eb80a5852ebb1c811b2b89d299b10573b", size = 1874920, upload-time = "2026-01-21T16:27:45.348Z" },
+    { url = "https://files.pythonhosted.org/packages/51/f8/c0e1ef27c66e15406fece94930e7d6feee4cb6374bbc02d945a630d6426e/torchvision-0.25.0-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:b75deafa2dfea3e2c2a525559b04783515e3463f6e830cb71de0fb7ea36fe233", size = 2344556, upload-time = "2026-01-21T16:27:40.125Z" },
+    { url = "https://files.pythonhosted.org/packages/68/2f/f24b039169db474e8688f649377de082a965fbf85daf4e46c44412f1d15a/torchvision-0.25.0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:f25aa9e380865b11ea6e9d99d84df86b9cc959f1a007cd966fc6f1ab2ed0e248", size = 8072351, upload-time = "2026-01-21T16:27:21.074Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/16/8f650c2e288977cf0f8f85184b90ee56ed170a4919347fc74ee99286ed6f/torchvision-0.25.0-cp312-cp312-win_amd64.whl", hash = "sha256:f9c55ae8d673ab493325d1267cbd285bb94d56f99626c00ac4644de32a59ede3", size = 4303059, upload-time = "2026-01-21T16:27:11.08Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/5b/1562a04a6a5a4cf8cf40016a0cdeda91ede75d6962cff7f809a85ae966a5/torchvision-0.25.0-cp313-cp313-macosx_12_0_arm64.whl", hash = "sha256:24e11199e4d84ba9c5ee7825ebdf1cd37ce8deec225117f10243cae984ced3ec", size = 1874918, upload-time = "2026-01-21T16:27:39.02Z" },
+    { url = "https://files.pythonhosted.org/packages/36/b1/3d6c42f62c272ce34fcce609bb8939bdf873dab5f1b798fd4e880255f129/torchvision-0.25.0-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:5f271136d2d2c0b7a24c5671795c6e4fd8da4e0ea98aeb1041f62bc04c4370ef", size = 2309106, upload-time = "2026-01-21T16:27:30.624Z" },
+    { url = "https://files.pythonhosted.org/packages/c7/60/59bb9c8b67cce356daeed4cb96a717caa4f69c9822f72e223a0eae7a9bd9/torchvision-0.25.0-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:855c0dc6d37f462482da7531c6788518baedca1e0847f3df42a911713acdfe52", size = 8071522, upload-time = "2026-01-21T16:27:29.392Z" },
+    { url = "https://files.pythonhosted.org/packages/32/a5/9a9b1de0720f884ea50dbf9acb22cbe5312e51d7b8c4ac6ba9b51efd9bba/torchvision-0.25.0-cp313-cp313-win_amd64.whl", hash = "sha256:cef0196be31be421f6f462d1e9da1101be7332d91984caa6f8022e6c78a5877f", size = 4321911, upload-time = "2026-01-21T16:27:35.195Z" },
+    { url = "https://files.pythonhosted.org/packages/52/99/dca81ed21ebaeff2b67cc9f815a20fdaa418b69f5f9ea4c6ed71721470db/torchvision-0.25.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:a8f8061284395ce31bcd460f2169013382ccf411148ceb2ee38e718e9860f5a7", size = 1896209, upload-time = "2026-01-21T16:27:32.159Z" },
+    { url = "https://files.pythonhosted.org/packages/28/cc/2103149761fdb4eaed58a53e8437b2d716d48f05174fab1d9fcf1e2a2244/torchvision-0.25.0-cp313-cp313t-manylinux_2_28_aarch64.whl", hash = "sha256:146d02c9876858420adf41f3189fe90e3d6a409cbfa65454c09f25fb33bf7266", size = 2310735, upload-time = "2026-01-21T16:27:22.327Z" },
+    { url = "https://files.pythonhosted.org/packages/76/ad/f4c985ad52ddd3b22711c588501be1b330adaeaf6850317f66751711b78c/torchvision-0.25.0-cp313-cp313t-manylinux_2_28_x86_64.whl", hash = "sha256:c4d395cb2c4a2712f6eb93a34476cdf7aae74bb6ea2ea1917f858e96344b00aa", size = 8089557, upload-time = "2026-01-21T16:27:27.666Z" },
+    { url = "https://files.pythonhosted.org/packages/63/cc/0ea68b5802e5e3c31f44b307e74947bad5a38cc655231d845534ed50ddb8/torchvision-0.25.0-cp313-cp313t-win_amd64.whl", hash = "sha256:5e6b449e9fa7d642142c0e27c41e5a43b508d57ed8e79b7c0a0c28652da8678c", size = 4344260, upload-time = "2026-01-21T16:27:17.018Z" },
 ]
 
 [[package]]
@@ -4029,6 +4239,18 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/91/88/b55b3117287a8540b76dbdd87733808d4d01c8067a3b339408c250bb3600/typeguard-4.5.1-py3-none-any.whl", hash = "sha256:44d2bf329d49a244110a090b55f5f91aa82d9a9834ebfd30bcc73651e4a8cc40", size = 36745, upload-time = "2026-02-19T16:09:01.6Z" },
 ]
 
+[[package]]
+name = "triton"
+version = "3.6.0"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/8c/f7/f1c9d3424ab199ac53c2da567b859bcddbb9c9e7154805119f8bd95ec36f/triton-3.6.0-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a6550fae429e0667e397e5de64b332d1e5695b73650ee75a6146e2e902770bea", size = 188105201, upload-time = "2026-01-20T16:00:29.272Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/12/b05ba554d2c623bffa59922b94b0775673de251f468a9609bc9e45de95e9/triton-3.6.0-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:e8e323d608e3a9bfcc2d9efcc90ceefb764a82b99dea12a86d643c72539ad5d3", size = 188214640, upload-time = "2026-01-20T16:00:35.869Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/a8/cdf8b3e4c98132f965f88c2313a4b493266832ad47fb52f23d14d4f86bb5/triton-3.6.0-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:74caf5e34b66d9f3a429af689c1c7128daba1d8208df60e81106b115c00d6fca", size = 188266850, upload-time = "2026-01-20T16:00:43.041Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/0b/37d991d8c130ce81a8728ae3c25b6e60935838e9be1b58791f5997b24a54/triton-3.6.0-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:10c7f76c6e72d2ef08df639e3d0d30729112f47a56b0c81672edc05ee5116ac9", size = 188289450, upload-time = "2026-01-20T16:00:49.136Z" },
+    { url = "https://files.pythonhosted.org/packages/35/f8/9c66bfc55361ec6d0e4040a0337fb5924ceb23de4648b8a81ae9d33b2b38/triton-3.6.0-cp313-cp313t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d002e07d7180fd65e622134fbd980c9a3d4211fb85224b56a0a0efbd422ab72f", size = 188400296, upload-time = "2026-01-20T16:00:56.042Z" },
+]
+
 [[package]]
 name = "typer"
 version = "0.24.1"