riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-31 02:02:21 +06:00
Code Issues Actions 23 Packages Projects Releases Wiki Activity

[Docs] Update compressed_tensors.md (#33961)

Browse Source 
* Update compressed_tensors.md

Fix some unfinished sections

* Update docs/source/en/quantization/compressed_tensors.md

Co-authored-by: Xiao Yuan <yuanx749@gmail.com>

---------

Co-authored-by: Xiao Yuan <yuanx749@gmail.com>
This commit is contained in:
Michael Goin 2024-10-10 09:22:41 -04:00 committed by GitHub
parent 4a3f1a686f
commit b2f09fb90f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
docs/source/en/quantization/compressed_tensors.md
View File

@ -19,15 +19,12 @@ The [`compressed-tensors`](https://github.com/neuralmagic/compressed-tensors) li
Some of the supported formats include:
1. `dense`
2. `int-quantized`: INT8 quantized models
- sample [model/config](https://huggingface.co/nm-testing/tinyllama-w8a8-compressed-hf-quantizer)
3. `float-quantized`: FP8 quantized models; currently support E4M3
- sample [model/config](https://huggingface.co/nm-testing/Meta-Llama-3-8B-Instruct-fp8-hf_compat/tree/main)
4. `pack-quantized`: INT4 or INT8 weight-quantized models, packed into INT32. For INT4, the weights have an INT4 range but are stored as INT8 and then packed into INT32.
- sample [model/config](nm-testing/tinyllama-w4a16-compressed-hf-quantizer)
2. `int-quantized` ([sample](https://huggingface.co/nm-testing/tinyllama-w8a8-compressed-hf-quantizer)): INT8 quantized models
3. `float-quantized` ([sample](https://huggingface.co/nm-testing/Meta-Llama-3-8B-Instruct-fp8-hf_compat)): FP8 quantized models; currently support E4M3
4. `pack-quantized` ([sample](https://huggingface.co/nm-testing/tinyllama-w4a16-compressed-hf-quantizer)): INT4 or INT8 weight-quantized models, packed into INT32. For INT4, the weights have an INT4 range but are stored as INT8 and then packed into INT32.
Compressed models can be easily created using [llm-compressor](https://github.com/vllm-project/llm-compressor).
Alternatively models can be created indepedenty and serialized with a compressed tensors config.
Alternatively models can be created independently and serialized with a compressed tensors config.
To find existing models on the Hugging Face Model Hub, search for the [`compressed-tensors` tag](https://huggingface.co/models?other=compressed-tensors).
@ -35,7 +32,7 @@ To find existing models on the Hugging Face Model Hub, search for the [`compress
- Weight and activation precisions: FP8, INT4, INT8 (for Q/DQ arbitrary precision is allowed for INT)
- Quantization scales and zero-points strategies: [tensor, channel, group, block, token](https://github.com/neuralmagic/compressed-tensors/blob/83b2e7a969d70606421a76b9a3d112646077c8de/src/compressed_tensors/quantization/quant_args.py#L43-L52)
- Dynamic per-token activation quantization (or any static strategy)
- Sparsity can be
- Sparsity in weights (unstructured or semi-structured like 2:4) can be composed with quantization for extreme compression
- Supports quantization of arbitrary modules, not just Linear modules
- Targeted support or ignoring of modules by name or class