|
--- |
|
base_model: https://huggingface.co/Phind/Phind-CodeLlama-34B-v2 |
|
inference: false |
|
license: llama2 |
|
model_creator: https://huggingface.co/Phind |
|
model_name: Phind-Codellama-34B-v2 |
|
model_type: llama |
|
quantized_by: latimar |
|
--- |
|
|
|
# Phind-CodeLlama-34B-v2 EXL2 |
|
|
|
Weights of [Phind-CodeLlama-34B-v2](https://huggingface.co/Phind/Phind-CodeLlama-34B-v2) converted |
|
to [EXL2](https://github.com/turboderp/exllamav2#exl2-quantization) format. |
|
|
|
Converted with the ExllamaV2 [convert.py](https://github.com/turboderp/exllamav2/blob/master/convert.py) script, |
|
exllamav2 [commit](https://github.com/turboderp/exllamav2/commit/31f31e1b08eeccf4a5ab31fd202ef3100dce8d22) |
|
|
|
|
|
| BPW (hb=8) | Human-Eval | Evol-Ins PPL | Wiki PPL | File Size (Gb) | |
|
| ----------- | ----------- | ------------ | ---------- | -------------- | |
|
| 2.55 | 0.402439 | 2.0944 | 18.9843 | 10.62 | |
|
| 3.0 | 0.664634 | 2.0600 | 11.2096 | 12.36 | |
|
| 4.625 | 0.701219 | 2.0401 | 6.7243 | 18.63 | |
|
| 5.0 | 0.670731 | 2.0391 | 6.6956 | 20.09 | |
|
|
|
## Datasets used for calibration and PPL measurement |
|
|
|
* [Calibration](https://huggingface.co/datasets/rombodawg/2XUNCENSORED_MegaCodeTraining188k) |
|
* [Wiki](https://huggingface.co/datasets/wikitext/blob/refs%2Fconvert%2Fparquet/wikitext-2-v1/validation/0000.parquet) |
|
* [Evol-Ins](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1/blob/refs%2Fconvert%2Fparquet/default/train/0000.parquet) |
|
|
|
|
|
### Conversion |
|
|
|
Conversion arguments: |
|
|
|
``` |
|
convert.py -i ${MODEL_DIR_FP16} -o ${WIP_DIR} -cf ${MODEL_DIR_EXL} -c ${CALIBRATION_DATASET} -r 200 -mr 32 -l 4096 -ml 4096 -hb 8 -b ${BPW} |
|
``` |
|
|
|
`2.55` quant was converted using even more raws: `-r 400 -mr 64` |
|
|
|
### Perplexity |
|
|
|
Perplexity was measured with [test_inference.py](https://github.com/turboderp/exllamav2/blob/master/test_inference.py) script: |
|
``` |
|
test_inference.py -m ${MODEL_DIR_EXL} -ed ${PPL_DATASET} |
|
``` |
|
|
|
### Human-Eval |
|
|
|
For the point of reference, Phind says that the original model achieves **73.8** Human-Eval score. |
|
|
|
Unfortunately, FP16/INT8 weights of this model won't fit on my RTX 4090, but FP16 quantized to NF4 fits, |
|
so I generated samples with [tf.human-eval.py](https://github.com/epicfilemcnulty/llm-tools/blob/main/eval/tf.human-eval.py) script: |
|
``` |
|
python tf.human-eval.py -m ${MODEL_DIR_FP16} -o nf4-samples.jsonl |
|
``` |
|
|
|
NF4 variant gives us **0.70731707** |
|
|
|
Samples for the Human-Eval scores of EXL2 quants were generated with [exl2.human-eval.py](https://github.com/epicfilemcnulty/llm-tools/blob/main/eval/exl2.human-eval.py) |
|
script like this: |
|
``` |
|
python exl2.human-eval.py -m ${MODEL_DIR_EXL2} -c 4096 ${BPW}-samples.jsonl |
|
``` |
|
|
