[#9463][feat] Add revision option to trtllm commands

tensorrt_llm/bench/benchmark/low_latency.py

@@ -225,7 +225,7 @@ def latency_command(
     if options.backend and options.backend.lower(
     ) in ALL_SUPPORTED_BACKENDS and options.backend.lower() != "tensorrt":
         if bench_env.checkpoint_path is None:
-            snapshot_download(options.model)
+            snapshot_download(options.model, revision=bench_env.revision)
 
         exec_settings = get_settings(params, metadata, bench_env.model,
                                      bench_env.checkpoint_path)
@@ -250,6 +250,7 @@ def latency_command(
             param_hint="backend")
 
     exec_settings["model"] = options.model
+    exec_settings["revision"] = bench_env.revision
     engine_tokens = exec_settings["settings_config"]["max_num_tokens"]
 
     # Update configuration with runtime options

tensorrt_llm/bench/benchmark/throughput.py

@@ -350,7 +350,7 @@ def throughput_command(
         # If we're dealing with a model name, perform a snapshot download to
         # make sure we have a local copy of the model.
         if bench_env.checkpoint_path is None:
-            snapshot_download(options.model)
+            snapshot_download(options.model, revision=bench_env.revision)
 
         exec_settings = get_settings(params, metadata, bench_env.model,
                                      bench_env.checkpoint_path)
@@ -376,6 +376,7 @@ def throughput_command(
             param_hint="backend")
 
     exec_settings["model"] = options.model
+    exec_settings["revision"] = bench_env.revision
     engine_bs = exec_settings["settings_config"]["max_batch_size"]
     engine_tokens = exec_settings["settings_config"]["max_num_tokens"]
 

tensorrt_llm/bench/dataclasses/configuration.py

@@ -25,6 +25,7 @@ class RuntimeConfig(BaseModel):
     model: str
     model_path: Optional[Path] = None
     engine_dir: Optional[Path] = None
+    revision: Optional[str] = None
     sw_version: str
     settings_config: ExecutorSettingsConfig
     # TODO: this is a dict corresponding to the Mapping class, the type should be

tensorrt_llm/bench/dataclasses/general.py

@@ -14,6 +14,7 @@ class BenchmarkEnvironment(BaseModel):
     model: str
     checkpoint_path: Optional[Path]
     workspace: Path
+    revision: Optional[str] = None
 
 
 class InferenceRequest(BaseModel):

tensorrt_llm/bench/dataclasses/reporting.py

@@ -306,6 +306,7 @@ def get_statistics_dict(self) -> Dict[str, Any]:
                 "model": self.rt_cfg.model,
                 "model_path": str(self.rt_cfg.model_path),
                 "engine_dir": str(self.rt_cfg.engine_dir),
+                "revision": self.rt_cfg.revision,
                 "version": self.rt_cfg.sw_version,
             },
         }
@@ -539,6 +540,7 @@ def report_statistics(self) -> None:
                 "===========================================================\n"
                 f"Model:\t\t\t{engine['model']}\n"
                 f"Model Path:\t\t{engine['model_path']}\n"
+                f"Revision:\t\t{engine['revision'] or 'N/A'}\n"
                 f"Engine Directory:\t{engine['engine_dir']}\n"
                 f"TensorRT LLM Version:\t{engine['version']}\n"
                 f"Dtype:\t\t\t{pretrain_cfg['dtype']}\n"
@@ -554,6 +556,7 @@ def report_statistics(self) -> None:
                 "===========================================================\n"
                 f"Model:\t\t\t{engine['model']}\n"
                 f"Model Path:\t\t{engine['model_path']}\n"
+                f"Revision:\t\t{engine['revision'] or 'N/A'}\n"
                 f"TensorRT LLM Version:\t{engine['version']}\n"
                 f"Dtype:\t\t\t{engine['dtype']}\n"
                 f"KV Cache Dtype:\t\t{engine['kv_cache_dtype']}\n"

tensorrt_llm/bench/utils/data.py

@@ -89,21 +89,21 @@ def create_dataset_from_stream(
     while (line := stream.readline()) and len(task_ids) < max_requests:
         # We expect the data to come in as a JSON string.
         # For example:
-        # {"prompt": "Generate an infinite response to the following:
+        # {"task_id": 1, "prompt": "Generate an infinite response to the following:
         # There once was a man who.", "output_tokens": 1000}
         #
         # For multimodal data, the data should be of the form:
-        # {"prompt": "Generate an infinite response to the following:
+        # {"task_id": 1, "prompt": "Generate an infinite response to the following:
         # There once was a man who.", "output_tokens": 1000,
         # "media_paths": ["/path/to/image1.jpg", "/path/to/image2.jpg"]}
         #
         # For LoRA data, the data should be of the form:
-        # {"prompt": "Generate an infinite response to the following:
+        # {"task_id": 1, "prompt": "Generate an infinite response to the following:
         # There once was a man who.", "output_tokens": 1000,
         # "lora_request": {"lora_name": "my_lora", "lora_int_id": 1, "lora_path": "/path/to/lora"}}
         #
         # Each line should be a complete JSON dictionary with no indentation
-        # or newline characters.
+        # or newline characters. The task_id field is required.
         data = json.loads(line)
         prompts.append(data.get("prompt"))
         media_paths.append(data.get("media_paths", None))

tensorrt_llm/commands/bench.py

@@ -1,4 +1,5 @@
 from pathlib import Path
+from typing import Optional
 
 import click
 
@@ -37,18 +38,25 @@
               type=click.Choice(severity_map.keys()),
               default='info',
               help="The logging level.")
+@click.option("--revision",
+              type=str,
+              default=None,
+              help="The revision to use for the HuggingFace model "
+              "(branch name, tag name, or commit id).")
 @click.pass_context
 def main(
     ctx,
     model: str,
     model_path: Path,
     workspace: Path,
     log_level: str,
+    revision: Optional[str],
 ) -> None:
     logger.set_level(log_level)
     ctx.obj = BenchmarkEnvironment(model=model,
                                    checkpoint_path=model_path,
-                                   workspace=workspace)
+                                   workspace=workspace,
+                                   revision=revision)
 
     # Create the workspace where we plan to store intermediate files.
     ctx.obj.workspace.mkdir(parents=True, exist_ok=True)

tensorrt_llm/commands/eval.py

@@ -92,6 +92,11 @@
               is_flag=True,
               default=False,
               help="Flag for HF transformers.")
+@click.option("--revision",
+              type=str,
+              default=None,
+              help="The revision to use for the HuggingFace model "
+              "(branch name, tag name, or commit id).")
 @click.option("--extra_llm_api_options",
               type=str,
               default=None,
@@ -106,7 +111,8 @@ def main(ctx, model: str, tokenizer: Optional[str], log_level: str,
          max_num_tokens: int, max_seq_len: int, tp_size: int, pp_size: int,
          ep_size: Optional[int], gpus_per_node: Optional[int],
          kv_cache_free_gpu_memory_fraction: float, trust_remote_code: bool,
-         extra_llm_api_options: Optional[str], disable_kv_cache_reuse: bool):
+         revision: Optional[str], extra_llm_api_options: Optional[str],
+         disable_kv_cache_reuse: bool):
     logger.set_level(log_level)
     build_config = BuildConfig(max_batch_size=max_batch_size,
                                max_num_tokens=max_num_tokens,
@@ -125,6 +131,7 @@ def main(ctx, model: str, tokenizer: Optional[str], log_level: str,
         "moe_expert_parallel_size": ep_size,
         "gpus_per_node": gpus_per_node,
         "trust_remote_code": trust_remote_code,
+        "revision": revision,
         "build_config": build_config,
         "kv_cache_config": kv_cache_config,
     }

tensorrt_llm/commands/serve.py

@@ -95,6 +95,7 @@ def get_llm_args(
         free_gpu_memory_fraction: float = 0.9,
         num_postprocess_workers: int = 0,
         trust_remote_code: bool = False,
+        revision: Optional[str] = None,
         reasoning_parser: Optional[str] = None,
         fail_fast_on_attention_window_too_large: bool = False,
         otlp_traces_endpoint: Optional[str] = None,
@@ -129,6 +130,7 @@ def get_llm_args(
         "moe_expert_parallel_size": moe_expert_parallel_size,
         "gpus_per_node": gpus_per_node,
         "trust_remote_code": trust_remote_code,
+        "revision": revision,
         "build_config": build_config,
         "max_batch_size": max_batch_size,
         "max_num_tokens": max_num_tokens,
@@ -317,6 +319,11 @@ def convert(self, value: Any, param: Optional["click.Parameter"],
               is_flag=True,
               default=False,
               help="Flag for HF transformers.")
+@click.option("--revision",
+              type=str,
+              default=None,
+              help="The revision to use for the HuggingFace model "
+              "(branch name, tag name, or commit id).")
 @click.option(
     "--extra_llm_api_options",
     type=str,
@@ -381,9 +388,9 @@ def serve(
         ep_size: Optional[int], cluster_size: Optional[int],
         gpus_per_node: Optional[int], kv_cache_free_gpu_memory_fraction: float,
         num_postprocess_workers: int, trust_remote_code: bool,
-        extra_llm_api_options: Optional[str], reasoning_parser: Optional[str],
-        tool_parser: Optional[str], metadata_server_config_file: Optional[str],
-        server_role: Optional[str],
+        revision: Optional[str], extra_llm_api_options: Optional[str],
+        reasoning_parser: Optional[str], tool_parser: Optional[str],
+        metadata_server_config_file: Optional[str], server_role: Optional[str],
         fail_fast_on_attention_window_too_large: bool,
         otlp_traces_endpoint: Optional[str], enable_chunked_prefill: bool,
         disagg_cluster_uri: Optional[str], media_io_kwargs: Optional[str],
@@ -418,6 +425,7 @@ def serve(
         free_gpu_memory_fraction=kv_cache_free_gpu_memory_fraction,
         num_postprocess_workers=num_postprocess_workers,
         trust_remote_code=trust_remote_code,
+        revision=revision,
         reasoning_parser=reasoning_parser,
         fail_fast_on_attention_window_too_large=
         fail_fast_on_attention_window_too_large,