Instructions to use arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0 with Transformers:

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

pipe = pipeline("text-generation", model="arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0

SGLang

How to use arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0 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 "arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0" \
    --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": "arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0",
		"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 "arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0" \
        --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": "arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0 with Docker Model Runner:
```
docker model run hf.co/arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0
```

arianraje commited on 10 days ago

Commit

4695f93

verified ·

1 Parent(s): 76aaf68

model card: results table refresh

Browse files

Files changed (1) hide show

README.md +2 -2

README.md CHANGED Viewed

@@ -95,7 +95,7 @@ evaluation** (below). Both arms are published; bias 1 is now the default.
 |---|---|---|---|---|
 | stage 1: val mean rel-MSE / worst layer / wikitext2 PPL | 0.00397 / L0 0.0371 / 26.54 | **0.00376** / L0 0.0357 / **25.35** | 0.0208 / - / - | 0.0175 / - / - |
 | stage 2a (4k): val KL / top1 agreement / PPL | 0.1297 / 0.8580 / 12.49 | **0.1258** / **0.8601** / 12.41 | 0.1437 / 0.8515 / 12.83 | 0.1819 / 0.8295 / 12.60 |
-| stage 2b (32k): val KL / top1 agreement / PPL | 0.2021 / 0.8056 / 11.93 | in flight | 0.2177 / 0.7985 / - | 0.2336 / 0.7875 / - |
 Val KL is the forward KL to the teacher over the full vocabulary on held-out
 packed text at the stage's sequence length; PPL is wikitext-2 at 4096 context.
@@ -105,5 +105,5 @@ step 20 of 287.
 **Not yet evaluated.** No downstream battery (commonsense, RULER
 needle-in-a-haystack, GSM8K / MATH / AIME, MMLU) has been run on any Mamba-3
 checkpoint; the numbers above are the trainers' own validation metrics.
-Stage 2b (32k KD) is complete for the bias-0 arm and in flight for the other. Stage 3 (on-policy distillation) has not been run.
 Intermediate checkpoints, published for the ladder.

 |---|---|---|---|---|
 | stage 1: val mean rel-MSE / worst layer / wikitext2 PPL | 0.00397 / L0 0.0371 / 26.54 | **0.00376** / L0 0.0357 / **25.35** | 0.0208 / - / - | 0.0175 / - / - |
 | stage 2a (4k): val KL / top1 agreement / PPL | 0.1297 / 0.8580 / 12.49 | **0.1258** / **0.8601** / 12.41 | 0.1437 / 0.8515 / 12.83 | 0.1819 / 0.8295 / 12.60 |
+| stage 2b (32k): val KL / top1 agreement / PPL | 0.2021 / 0.8056 / 11.93 | 0.1957 / 0.8089 / 11.81 | 0.2177 / 0.7985 / - | 0.2336 / 0.7875 / - |
 Val KL is the forward KL to the teacher over the full vocabulary on held-out
 packed text at the stage's sequence length; PPL is wikitext-2 at 4096 context.
 **Not yet evaluated.** No downstream battery (commonsense, RULER
 needle-in-a-haystack, GSM8K / MATH / AIME, MMLU) has been run on any Mamba-3
 checkpoint; the numbers above are the trainers' own validation metrics.
+Stage 2b (32k KD) is complete for both arms. Stage 3 (on-policy distillation) has not been run.
 Intermediate checkpoints, published for the ladder.