Text Generation
Transformers
Safetensors
English
cloverlm
causal-lm
quartet-ii
nvfp4
low-precision-training
pretrained
custom_code
Instructions to use daslab-testing/CloverLM with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use daslab-testing/CloverLM with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="daslab-testing/CloverLM", trust_remote_code=True)# Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("daslab-testing/CloverLM", trust_remote_code=True, dtype="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps
- vLLM
How to use daslab-testing/CloverLM with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "daslab-testing/CloverLM" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "daslab-testing/CloverLM", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/daslab-testing/CloverLM
- SGLang
How to use daslab-testing/CloverLM with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "daslab-testing/CloverLM" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "daslab-testing/CloverLM", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "daslab-testing/CloverLM" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "daslab-testing/CloverLM", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use daslab-testing/CloverLM with Docker Model Runner:
docker model run hf.co/daslab-testing/CloverLM
Update vllm_plugin/quartet2_quant.py
Browse files
vllm_plugin/quartet2_quant.py
CHANGED
|
@@ -93,9 +93,9 @@ class QuartetIILinearMethod(LinearMethodBase):
|
|
| 93 |
|
| 94 |
weight = layer.weight
|
| 95 |
orig_shape = x.shape
|
|
|
|
| 96 |
flat_x = x.reshape(-1, x.shape[-1])
|
| 97 |
|
| 98 |
-
# Quartet II requires rows to be multiples of 128; pad if needed.
|
| 99 |
num_rows = flat_x.shape[0]
|
| 100 |
remainder = num_rows % 128
|
| 101 |
if remainder != 0:
|
|
@@ -104,6 +104,13 @@ class QuartetIILinearMethod(LinearMethodBase):
|
|
| 104 |
else:
|
| 105 |
pad_rows = 0
|
| 106 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 107 |
input_amax = abs_max(flat_x)
|
| 108 |
weight_amax = abs_max(weight)
|
| 109 |
|
|
@@ -128,6 +135,8 @@ class QuartetIILinearMethod(LinearMethodBase):
|
|
| 128 |
|
| 129 |
if pad_rows > 0:
|
| 130 |
output = output[:num_rows]
|
|
|
|
|
|
|
| 131 |
|
| 132 |
output = output.reshape(*orig_shape[:-1], output.shape[-1])
|
| 133 |
if bias is not None:
|
|
|
|
| 93 |
|
| 94 |
weight = layer.weight
|
| 95 |
orig_shape = x.shape
|
| 96 |
+
out_features = weight.shape[0]
|
| 97 |
flat_x = x.reshape(-1, x.shape[-1])
|
| 98 |
|
|
|
|
| 99 |
num_rows = flat_x.shape[0]
|
| 100 |
remainder = num_rows % 128
|
| 101 |
if remainder != 0:
|
|
|
|
| 104 |
else:
|
| 105 |
pad_rows = 0
|
| 106 |
|
| 107 |
+
w_remainder = out_features % 128
|
| 108 |
+
if w_remainder != 0:
|
| 109 |
+
w_pad = 128 - w_remainder
|
| 110 |
+
weight = F.pad(weight, (0, 0, 0, w_pad))
|
| 111 |
+
else:
|
| 112 |
+
w_pad = 0
|
| 113 |
+
|
| 114 |
input_amax = abs_max(flat_x)
|
| 115 |
weight_amax = abs_max(weight)
|
| 116 |
|
|
|
|
| 135 |
|
| 136 |
if pad_rows > 0:
|
| 137 |
output = output[:num_rows]
|
| 138 |
+
if w_pad > 0:
|
| 139 |
+
output = output[:, :out_features]
|
| 140 |
|
| 141 |
output = output.reshape(*orig_shape[:-1], output.shape[-1])
|
| 142 |
if bias is not None:
|