eval
Browse files
README.md
CHANGED
|
@@ -1 +1,53 @@
|
|
| 1 |
AI Model Name: Llama 3 8B "Built with Meta Llama 3" https://llama.meta.com/llama3/license/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
AI Model Name: Llama 3 8B "Built with Meta Llama 3" https://llama.meta.com/llama3/license/
|
| 2 |
+
|
| 3 |
+
Baseline evaluation results:
|
| 4 |
+
|
| 5 |
+
```
|
| 6 |
+
hf (pretrained=meta-llama/Meta-Llama-3-8B-Instruct), gen_kwargs: (None), limit: None, num_fewshot: None, batch_size: 16
|
| 7 |
+
| Tasks |Version|Filter|n-shot| Metric |Value | |Stderr|
|
| 8 |
+
|-------------|------:|------|-----:|--------|-----:|---|-----:|
|
| 9 |
+
|winogrande | 1|none | 0|acc |0.7198|± |0.0126|
|
| 10 |
+
|piqa | 1|none | 0|acc |0.7873|± |0.0095|
|
| 11 |
+
| | |none | 0|acc_norm|0.7867|± |0.0096|
|
| 12 |
+
|hellaswag | 1|none | 0|acc |0.5767|± |0.0049|
|
| 13 |
+
| | |none | 0|acc_norm|0.7585|± |0.0043|
|
| 14 |
+
|arc_easy | 1|none | 0|acc |0.8140|± |0.0080|
|
| 15 |
+
| | |none | 0|acc_norm|0.7971|± |0.0083|
|
| 16 |
+
|arc_challenge| 1|none | 0|acc |0.5290|± |0.0146|
|
| 17 |
+
| | |none | 0|acc_norm|0.5674|± |0.0145|
|
| 18 |
+
```
|
| 19 |
+
|
| 20 |
+
This repo evaluation results (AQLM with no global fine-tuning):
|
| 21 |
+
|
| 22 |
+
```
|
| 23 |
+
hf (pretrained=catid/cat-llama-3-8b-instruct-aqlm-noft), gen_kwargs: (None), limit: None, num_fewshot: None, batch_size: 16
|
| 24 |
+
| Tasks |Version|Filter|n-shot| Metric |Value | |Stderr|
|
| 25 |
+
|-------------|------:|------|-----:|--------|-----:|---|-----:|
|
| 26 |
+
|winogrande | 1|none | 0|acc |0.7119|± |0.0127|
|
| 27 |
+
|piqa | 1|none | 0|acc |0.7807|± |0.0097|
|
| 28 |
+
| | |none | 0|acc_norm|0.7824|± |0.0096|
|
| 29 |
+
|hellaswag | 1|none | 0|acc |0.5716|± |0.0049|
|
| 30 |
+
| | |none | 0|acc_norm|0.7539|± |0.0043|
|
| 31 |
+
|arc_easy | 1|none | 0|acc |0.8152|± |0.0080|
|
| 32 |
+
| | |none | 0|acc_norm|0.7866|± |0.0084|
|
| 33 |
+
|arc_challenge| 1|none | 0|acc |0.5043|± |0.0146|
|
| 34 |
+
| | |none | 0|acc_norm|0.5555|± |0.0145|
|
| 35 |
+
```
|
| 36 |
+
|
| 37 |
+
To reproduce evaluation results:
|
| 38 |
+
|
| 39 |
+
```bash
|
| 40 |
+
git clone https://github.com/EleutherAI/lm-evaluation-harness
|
| 41 |
+
cd lm-evaluation-harness
|
| 42 |
+
|
| 43 |
+
conda create -n lmeval python=3.10 -y && conda activate lmeval
|
| 44 |
+
pip install -e .
|
| 45 |
+
pip install accelerate aqlm"[gpu,cpu]"
|
| 46 |
+
|
| 47 |
+
accelerate launch lm_eval --model hf \
|
| 48 |
+
--model_args pretrained=catid/cat-llama-3-8b-instruct-aqlm-noft \
|
| 49 |
+
--tasks winogrande,piqa,hellaswag,arc_easy,arc_challenge \
|
| 50 |
+
--batch_size 16
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
You can run this model as a `transformers` model using https://github.com/oobabooga/text-generation-webui
|