Instructions to use crabby-devil/zils-lite-decision-r16e2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use crabby-devil/zils-lite-decision-r16e2 with Transformers:

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

pipe = pipeline("image-text-to-text", model="crabby-devil/zils-lite-decision-r16e2")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("crabby-devil/zils-lite-decision-r16e2")
model = AutoModelForMultimodalLM.from_pretrained("crabby-devil/zils-lite-decision-r16e2", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use crabby-devil/zils-lite-decision-r16e2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "crabby-devil/zils-lite-decision-r16e2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "crabby-devil/zils-lite-decision-r16e2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/crabby-devil/zils-lite-decision-r16e2

SGLang

How to use crabby-devil/zils-lite-decision-r16e2 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 "crabby-devil/zils-lite-decision-r16e2" \
    --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": "crabby-devil/zils-lite-decision-r16e2",
		"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 "crabby-devil/zils-lite-decision-r16e2" \
        --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": "crabby-devil/zils-lite-decision-r16e2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use crabby-devil/zils-lite-decision-r16e2 with Docker Model Runner:
```
docker model run hf.co/crabby-devil/zils-lite-decision-r16e2
```

d3-lite-lora-r16-lr2e4-e2-fix: d3-lite + LoRA through the Zils d3-lite chassis

What this is

A full Qwen3.5-0.8B checkpoint (language and vision tensors, bf16; no lm_head or mtp tensors) for the Zils (Bittensor SN127) d3-lite soup pool.

Start: vllm-sr/d3-lite @ 03560c26f5199c3c04129a2ccb38960c913635e0 (Apache-2.0), built on Qwen/Qwen3.5-0.8B.
Change: LoRA rank 16, alpha 32, on the 114 attention projections of the language model (q/k/v/o_proj; in_proj_qkv/z/b/a, out_proj), trained through the vllm-sr/d3-lite chassis (native prompts, frozen FP32 readout), merged in FP32 and stored in bf16. Embeddings, norms, MLPs and the vision tower are the start's.
Body sha256 (zils.bodies.body_hash): e4f0cbcf3130cce94a8cee09f75a923b6cfdf3eab676a16c2c872d3f115bff20
Relative change from Qwen/Qwen3.5-0.8B @ 2fc06364715b over the language layers: 0.02264

Training

12851 training rows (arts 2916, language 2489, reasoning 2524, retrieval 2246, tools 2676); 767 optimizer steps of 16 rows, peak lr 0.0002, seed 11.
No training or selection row shares an id, a source row or a state text with the Zils public seed files (checked at load; 0 rows removed).
Distillation anchor: the rows of four families (retrieval-known-score-k2, arts-known-noul-k5, reasoning-known-choice-k3, arts-known-score-k3) were trained on the start model's own answer probabilities instead of the generated labels.
Checkpoint: step 308, chosen on the merged-weights dev score.

Licence

Apache-2.0: see LICENSE and NOTICE (the start bodies' notices are kept and the modification is stated).

Use

Zils validators grade it as a soup body (zils.soup.build on the pinned chassis, then zils.d3.DecisionModel). config.json is Qwen/Qwen3.5-0.8B's (tied embeddings), so transformers also loads it as Qwen3_5ForConditionalGeneration.

Downloads last month: 25

Safetensors

Model size

0.9B params

Tensor type

BF16

Model tree for crabby-devil/zils-lite-decision-r16e2

Base model

Qwen/Qwen3.5-0.8B-Base

Finetuned

Qwen/Qwen3.5-0.8B

Finetuned

vllm-sr/d3-lite

Finetuned

(18)

this model