Instructions to use mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1

SGLang

How to use mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1 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 "mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1 with Docker Model Runner:
```
docker model run hf.co/mrs83/Kurtis-EON1-Hybrid-0.7B-v0.1.1
```

mrs83 commited on 20 days ago

Commit

6b3235f

verified ·

1 Parent(s): 4737c54

Upload modeling_hybrid.py

Browse files

Files changed (1) hide show

modeling_hybrid.py +6 -1

modeling_hybrid.py CHANGED Viewed

@@ -215,7 +215,12 @@ class HybridEchoModel(Qwen2PreTrainedModel):
             if past_key_values is None:
                 past_key_values = HybridEchoCache(config=self.config) if use_dsrn_cache else None
-        self._dsrn_input_states = dsrn_states  # ALWAYS run injectors regardless of cache mode.
         # use_dsrn_cache only gates whether output states are *returned* in the
         # HybridEchoCache.  Setting this to [] when use_cache=False (training mode)
         # caused the hook to exit early → injectors bypassed → grad_norm=0.

             if past_key_values is None:
                 past_key_values = HybridEchoCache(config=self.config) if use_dsrn_cache else None
+        # Detach DSRN states before carrying forward to the next step.
+        # Without .detach(), _dsrn_input_states holds tensors with grad_fn
+        # from the previous forward pass.  Step N+1's graph then includes
+        # step N's graph as a parent, preventing step N's activation tensors
+        # from being freed after backward — accumulating ~11.78 GiB per step.
+        self._dsrn_input_states = [(h.detach(), c.detach()) for h, c in dsrn_states]
         # use_dsrn_cache only gates whether output states are *returned* in the
         # HybridEchoCache.  Setting this to [] when use_cache=False (training mode)
         # caused the hook to exit early → injectors bypassed → grad_norm=0.