Skip to content
This repository has been archived by the owner on Aug 30, 2024. It is now read-only.

[Neural Speed] Fix Baichuan, chatGLM1&2&3 acc issue #285

Merged
merged 1 commit into from
Jun 7, 2024
Merged

Conversation

zhentaoyu
Copy link
Contributor

@zhentaoyu zhentaoyu commented Jun 6, 2024

Type of Change

feature or bug fix or documentation or others
API changed or not

Description

detail description
Issues: xxx

due to logits_all memory copy

ne_baichuan_q_int4_bestla_cint8_sym_sfp32_g32.bin    
|    Tasks     |Version|Filter|n-shot|  Metric  |Value |   |Stderr|
|--------------|------:|------|-----:|----------|-----:|---|-----:|
|lambada_openai|      1|none  |     0|perplexity|4.0944|±  |0.1164|
|              |       |none  |     0|acc       |0.6662|±  |0.0066|

ne_chatglm3_q_int4_bestla_cint8_sym_sfp32_g32.bin   
|    Tasks     |Version|Filter|n-shot|  Metric  |Value|   |Stderr|
|--------------|------:|------|-----:|----------|----:|---|-----:|
|lambada_openai|      1|none  |     0|perplexity|9.339|±  |0.4476|
|              |       |none  |     0|acc       |0.596|±  |0.0068|

ne_chatglm2_q_int4_bestla_cint8_sym_sfp32_g32.bin
|    Tasks     |Version|Filter|n-shot|  Metric  | Value |   |Stderr|
|--------------|------:|------|-----:|----------|------:|---|-----:|
|lambada_openai|      1|none  |     0|perplexity|13.0181|±  |0.6216|
|              |       |none  |     0|acc       | 0.5263|±  |0.0070|

due to tokenizer

ne_chatglm_q_int4_bestla_cint8_sym_sfp32_g32.bin
|    Tasks     |Version|Filter|n-shot|  Metric  |  Value  |   | Stderr |
|--------------|------:|------|-----:|----------|--------:|---|-------:|
|lambada_openai|      1|none  |     0|perplexity|1181.2761|±  |178.7728|
|              |       |none  |     0|acc       |   0.4194|±  |  0.0069|

|Tasks|Version|Filter|n-shot| Metric |Value|   |Stderr|
|-----|------:|------|-----:|--------|----:|---|-----:|
|piqa |      1|none  |     0|acc     |0.506|±  |0.0117|
|     |       |none  |     0|acc_norm|0.488|±  |0.0117|

|  Tasks   |Version|Filter|n-shot|Metric|Value |   |Stderr|
|----------|------:|------|-----:|------|-----:|---|-----:|
|winogrande|      1|none  |     0|acc   |0.4996|±  |0.0141|

Expected Behavior & Potential Risk

the expected behavior that triggered by this PR

How has this PR been tested?

how to reproduce the test (including hardware information)

Dependency Change?

any library dependency introduced or removed

@zhentaoyu zhentaoyu requested review from intellinjun, Zhenzhong1 and a32543254 and removed request for intellinjun June 6, 2024 06:00
Copy link
Contributor

@a32543254 a32543254 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@intellinjun
Copy link
Contributor

intellinjun commented Jun 6, 2024

@zhentaoyu zhentaoyu added the bug Something isn't working label Jun 6, 2024
@zhentaoyu
Copy link
Contributor Author

zhentaoyu commented Jun 6, 2024

https://inteltf-jenk.sh.intel.com/job/neural_speed_extension/170/

waiting for ext test results.

@zhentaoyu
Copy link
Contributor Author

ne_chatglm_f32.bin
|    Tasks     |Version|Filter|n-shot|  Metric  |  Value  |   | Stderr |
|--------------|------:|------|-----:|----------|--------:|---|-------:|
|lambada_openai|      1|none  |     0|perplexity|1089.0576|±  |166.4332|
|              |       |none  |     0|acc       |   0.4236|±  |  0.0069|

@a32543254 a32543254 merged commit ef42ce1 into main Jun 7, 2024
15 checks passed
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
bug Something isn't working ready to merge
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants