Update README.md
Browse files
README.md
CHANGED
@@ -21,12 +21,15 @@ pipeline_tag: text-generation
|
|
21 |
|
22 |
# Installation
|
23 |
```
|
24 |
-
pip install transformers
|
25 |
pip install --pre torchao --index-url https://download.pytorch.org/whl/nightly/cu126
|
26 |
-
pip install [email protected]:EleutherAI/lm-evaluation-harness.git
|
27 |
pip install vllm --pre --extra-index-url https://wheels.vllm.ai/nightly
|
28 |
```
|
29 |
|
|
|
|
|
|
|
|
|
30 |
# Quantization Recipe
|
31 |
We used following code to get the quantized model:
|
32 |
|
@@ -81,11 +84,6 @@ print(f"{save_to} model:", benchmark_fn(quantized_model.generate, **inputs, max_
|
|
81 |
# Model Quality
|
82 |
We rely on [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) to evaluate the quality of the quantized model.
|
83 |
|
84 |
-
## Installing the nightly version to get most recent updates
|
85 |
-
```
|
86 |
-
pip install git+https://github.com/EleutherAI/lm-evaluation-harness
|
87 |
-
```
|
88 |
-
|
89 |
## baseline
|
90 |
```
|
91 |
lm_eval --model hf --model_args pretrained=microsoft/Phi-4-mini-instruct --tasks hellaswag --device cuda:0 --batch_size 8
|
|
|
21 |
|
22 |
# Installation
|
23 |
```
|
24 |
+
pip install git+https://github.com/huggingface/transformers
|
25 |
pip install --pre torchao --index-url https://download.pytorch.org/whl/nightly/cu126
|
|
|
26 |
pip install vllm --pre --extra-index-url https://wheels.vllm.ai/nightly
|
27 |
```
|
28 |
|
29 |
+
Also need to install lm-eval from source:
|
30 |
+
https://github.com/EleutherAI/lm-evaluation-harness#install
|
31 |
+
|
32 |
+
|
33 |
# Quantization Recipe
|
34 |
We used following code to get the quantized model:
|
35 |
|
|
|
84 |
# Model Quality
|
85 |
We rely on [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) to evaluate the quality of the quantized model.
|
86 |
|
|
|
|
|
|
|
|
|
|
|
87 |
## baseline
|
88 |
```
|
89 |
lm_eval --model hf --model_args pretrained=microsoft/Phi-4-mini-instruct --tasks hellaswag --device cuda:0 --batch_size 8
|