Detect GigaChat3-10-A1.8B as deepseek lite#17420
Merged
CISC merged 2 commits intoggml-org:masterfrom Nov 21, 2025
Merged
Conversation
Contributor
Author
|
I added a test Q8_0 gguf here: https://huggingface.co/ubergarm/GigaChat3-10B-A1.8B-GGUF/tree/main |
Contributor
Author
|
Oops i need to get rid of an accidently commit file, will force push to fix. |
Hardcodes checking number of layers to detect if lite version of deepseek.
ubergarm
force-pushed
the
ug/gigachat3-lite
branch
from
3cddaab to
e1bfe51
Compare
Contributor
Author
|
Perplexity seems reasonable:
Always a bit funky when quants have lower perplexity than the original bf16... though it happens sometimes and in this case the values are very similar within the noise. |
CISC
approved these changes
Nov 21, 2025
Member
CISC
left a comment
CISC
left a comment
There was a problem hiding this comment.
Can you add a comment describing which models we are detecting in case this needs to be fine-tuned in the future?
deepseek lite variants include DeepSeek-V2-Lite, GigaChat3-10B-A1.8B
Member
|
Thank you! |
Anico2
added a commit
to Anico2/llama.cpp
that referenced
this pull request
Jan 15, 2026
* Detect GigaChat3-10-A1.8B as deepseek lite Hardcodes checking number of layers to detect if lite version of deepseek. * Add commnent identifying deepseek lite variants deepseek lite variants include DeepSeek-V2-Lite, GigaChat3-10B-A1.8B
blime4
referenced
this pull request
in blime4/llama.cpp
Feb 5, 2026
* Detect GigaChat3-10-A1.8B as deepseek lite Hardcodes checking number of layers to detect if lite version of deepseek. * Add commnent identifying deepseek lite variants deepseek lite variants include DeepSeek-V2-Lite, GigaChat3-10B-A1.8B
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Hardcodes checking number of layers to detect if a model is the lite version of deepseek.
Tested with bf16 and q8_0 version of GigaChat3-10B-A1.8B and discussed realizing it was a
liteversion similar to DeepSeek-V2-Lite. That model had 27 layers, but GigaChat3 has 26 and that is used to detect thelitevariant as discussed here: https://huggingface.co/ai-sage/GigaChat3-10B-A1.8B/discussions/1#691fb161ac024c8eb626ab36I'd like if anyone else could test. I'll update after testing perplexity to make sure the value looks sane. I haven't uploaded a gguf yet as the template has a parse error and wanted to get it updated before baking it in. That is discussed here: https://huggingface.co/ai-sage/GigaChat3-702B-A36B-preview-bf16/discussions/1