Text Generation
Transformers
Safetensors
English
Italian
quark
causal-lm
small-language-model
gqa
rope
swiglu
bash
code
custom_code
Instructions to use ThingAI/ARK-72M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use ThingAI/ARK-72M with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="ThingAI/ARK-72M", trust_remote_code=True)# Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("ThingAI/ARK-72M", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use ThingAI/ARK-72M with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "ThingAI/ARK-72M" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "ThingAI/ARK-72M", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/ThingAI/ARK-72M
- SGLang
How to use ThingAI/ARK-72M 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 "ThingAI/ARK-72M" \ --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": "ThingAI/ARK-72M", "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 "ThingAI/ARK-72M" \ --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": "ThingAI/ARK-72M", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use ThingAI/ARK-72M with Docker Model Runner:
docker model run hf.co/ThingAI/ARK-72M
fix: SDPA in float32 per stabilità numerica
Browse files- modeling_quark.py +4 -3
modeling_quark.py
CHANGED
|
@@ -78,13 +78,14 @@ class GroupedQueryAttention(nn.Module):
|
|
| 78 |
if self.n_groups > 1:
|
| 79 |
k = k.repeat_interleave(self.n_groups, dim=1)
|
| 80 |
v = v.repeat_interleave(self.n_groups, dim=1)
|
| 81 |
-
#
|
| 82 |
-
|
| 83 |
-
q, k = q.
|
| 84 |
out = F.scaled_dot_product_attention(
|
| 85 |
q, k, v, is_causal=True,
|
| 86 |
dropout_p=self.drop if self.training else 0.0,
|
| 87 |
)
|
|
|
|
| 88 |
out = out.transpose(1, 2).contiguous().view(B, T, self.n_heads * self.head_dim)
|
| 89 |
return self.o_proj(out)
|
| 90 |
|
|
|
|
| 78 |
if self.n_groups > 1:
|
| 79 |
k = k.repeat_interleave(self.n_groups, dim=1)
|
| 80 |
v = v.repeat_interleave(self.n_groups, dim=1)
|
| 81 |
+
# SDPA in float32 per stabilità numerica, poi riporta al dtype originale
|
| 82 |
+
orig_dtype = v.dtype
|
| 83 |
+
q, k, v = q.float(), k.float(), v.float()
|
| 84 |
out = F.scaled_dot_product_attention(
|
| 85 |
q, k, v, is_causal=True,
|
| 86 |
dropout_p=self.drop if self.training else 0.0,
|
| 87 |
)
|
| 88 |
+
out = out.to(orig_dtype)
|
| 89 |
out = out.transpose(1, 2).contiguous().view(B, T, self.n_heads * self.head_dim)
|
| 90 |
return self.o_proj(out)
|
| 91 |
|