Skip to content
This repository was archived by the owner on Oct 25, 2024. It is now read-only.
This repository was archived by the owner on Oct 25, 2024. It is now read-only.

After fine-tuning qwen2-1.5B-instruction and quantifying its AWQ, an error occurred while using Intel Extension for Transformers and CPU for inference. But when I used the same method to fine tune and quantify qwen1.5-4B chat before, I could use Intel Extension for Transformers to accelerate CPU inference. 对qwen2-1.5B-instruct微调并且awq量化后,使用intel-extension-for-transformers和CPU进行推理时出错。但我之前使用同样的方式微调和量化的qwen1.5-4B-chat时是可以使用intel-extension-for-transformers加速CPU推理的。 #1697

Open
@Autism-al

Description

@Autism-al

model.cpp: loading model from runtime_outs/ne_qwen2_q_autoround.bin
The number of ne_parameters is wrong.
init: n_vocab = 151936
init: n_embd = 1536
init: n_mult = 8960
init: n_head = 12
init: n_head_kv = 0
init: n_layer = 28
init: n_rot = 128
init: ftype = 0
init: max_seq_len= 32768
init: n_ff = 8960
init: n_parts = 1
MODEL_ASSERT: /root/w0/workspace/neuralspeed-wheel-build/nlp_repo/neural_speed/./models/qwen/qwen.h:48: false
/tmp/tmp9b4073w1: line 3: 55575 Aborted python /home/lmf/llm/Qwen2-finetuning/awq_intel_extension.py
ERROR conda.cli.main_run:execute(124): conda run python /home/lmf/llm/Qwen2-finetuning/awq_intel_extension.py failed. (See above for error)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions