Skip to content

尝试量化了下DeepSeek-R1-Distill-Llama-70B 但是测下来效果很差 请问怎么调比较好 #31

Description

@artetaout

用的这个命令

python3 examples/quant_model.py \
--model_path /sda/DeepSeek-R1-Distill-Llama-70B \
--tokenizer_path /sda/DeepSeek-R1-Distill-Llama-70B} \
--dtype float16 \
--smooth false \
--rotation true \
--dataset wikitext2 \
--nsamples 128 \
--w_quantizer FixedQuantize \
--w_group_size -1 \
--gptq_mse true \
--gptq_groupsize -1 \
--save_path DeepSeek-R1-Distill-Llama-70B-QQQ

用lm_eval来评测 效果很差

  • 原版
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value|   |Stderr|                                                                            |-----|------:|----------------|-----:|-----------|---|----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.934|±  |0.0068|                                                                            |     |       |strict-match    |     5|exact_match|↑  |0.928|±  |0.0071|
  • QQQ
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|                                                                           |-----|------:|----------------|-----:|-----------|---|-----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.0174|±  |0.0036|                                                                           |     |       |strict-match    |     5|exact_match|↑  |0.0000|±  |0.0000|

请问需要调哪些呢?有没有一些方法论呢?

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions