Instructions to use theCoderWithHat/mhc-qwen3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use theCoderWithHat/mhc-qwen3 with Transformers:

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

pipe = pipeline("text-generation", model="theCoderWithHat/mhc-qwen3")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import Qwen3MHCForCausalLMV2
model = Qwen3MHCForCausalLMV2.from_pretrained("theCoderWithHat/mhc-qwen3", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use theCoderWithHat/mhc-qwen3 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "theCoderWithHat/mhc-qwen3"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "theCoderWithHat/mhc-qwen3",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/theCoderWithHat/mhc-qwen3

SGLang

How to use theCoderWithHat/mhc-qwen3 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 "theCoderWithHat/mhc-qwen3" \
    --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": "theCoderWithHat/mhc-qwen3",
		"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 "theCoderWithHat/mhc-qwen3" \
        --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": "theCoderWithHat/mhc-qwen3",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use theCoderWithHat/mhc-qwen3 with Docker Model Runner:
```
docker model run hf.co/theCoderWithHat/mhc-qwen3
```

100k

by theCoderWithHat - opened 17 days ago

base: refs/heads/main

←

from: refs/pr/1

Discussion Files changed

-6

This PR is in draft mode

Files changed (3) hide show

README.md +3 -3
model.safetensors +1 -1
training_state.pt +2 -2

README.md CHANGED Viewed

@@ -9,13 +9,13 @@ pipeline_tag: text-generation
 # Qwen3 mHC
-This checkpoint is a Manifold-Constrained Hyper-Connections (mHC) V2 variant of Qwen/Qwen3-0.6B, trained for 100k steps in a parity-mixed setup. It is intended for research on residual stream mixing and hyper-connection behavior.
 ## Model Description
 - **Base model:** Qwen/Qwen3-0.6B
 - **Architecture:** Qwen3 with mHC V2 hyper-connections (stream-mixing)
-- **Checkpoint:** 100,000 steps
 - **Language(s):** Multilingual (see data notes)
 - **License:** Apache-2.0 (inherits base model license)
@@ -37,7 +37,7 @@ This checkpoint was trained on multilingual pretokenized datasets, primarily San
 ## Training Procedure
 - Converted from a Qwen3 base checkpoint into an mHC V2 model.
-- Trained for 100k steps in a parity-mixed run.
 - Uses Sinkhorn-based projection for residual mixing stability.
 ## Evaluation

 # Qwen3 mHC
+This checkpoint is a Manifold-Constrained Hyper-Connections (mHC) V2 variant of Qwen/Qwen3-0.6B, trained for 30k steps in a parity-mixed setup. It is intended for research on residual stream mixing and hyper-connection behavior.
 ## Model Description
 - **Base model:** Qwen/Qwen3-0.6B
 - **Architecture:** Qwen3 with mHC V2 hyper-connections (stream-mixing)
+- **Checkpoint:** 30,000 steps
 - **Language(s):** Multilingual (see data notes)
 - **License:** Apache-2.0 (inherits base model license)
 ## Training Procedure
 - Converted from a Qwen3 base checkpoint into an mHC V2 model.
+- Trained for 30k steps in a parity-mixed run.
 - Uses Sinkhorn-based projection for residual mixing stability.
 ## Evaluation

model.safetensors CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a53fe57c6e9db80d33eb308f3653bb20dbc38699ed096c973a3998e8ddd4c51
 size 1192998544

 version https://git-lfs.github.com/spec/v1
+oid sha256:aad3344026d91d0467a056746fa3bb2d3f085778ae0715a42731cb5247134d3e
 size 1192998544

training_state.pt CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:72b0d8f5cb125d37f2bd3181882114979ef1fda40beccc460b99e660e540aadc
-size 2387347844

 version https://git-lfs.github.com/spec/v1
+oid sha256:f2d23078ace67f6ac34dd3d8b7ea4846d4c0347d324e30a73ce2ab64e09d09dc
+size 2386717700