Instructions to use polygramme/PolyClerk-12B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use polygramme/PolyClerk-12B with Transformers:

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

pipe = pipeline("text-generation", model="polygramme/PolyClerk-12B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("polygramme/PolyClerk-12B")
model = AutoModelForCausalLM.from_pretrained("polygramme/PolyClerk-12B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use polygramme/PolyClerk-12B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "polygramme/PolyClerk-12B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "polygramme/PolyClerk-12B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/polygramme/PolyClerk-12B

SGLang

How to use polygramme/PolyClerk-12B 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 "polygramme/PolyClerk-12B" \
    --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": "polygramme/PolyClerk-12B",
		"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 "polygramme/PolyClerk-12B" \
        --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": "polygramme/PolyClerk-12B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use polygramme/PolyClerk-12B with Docker Model Runner:
```
docker model run hf.co/polygramme/PolyClerk-12B
```

PolyClerk-12B

File size: 3,098 Bytes

---
license: mit
base_model: zai-org/GLM-4.5-Air
tags:
  - legal
  - contract-review
  - document-analysis
  - lora
  - glm
library_name: transformers
pipeline_tag: text-generation
---

# PolyClerk-12B

PolyClerk-12B (12B active / 106B total MoE parameters, GLM-4.5-Air derivative) is fine-tuned for legal work product: counterparty markup
analysis, redlining, contract drafting, and clause-level document review over
very long contexts (up to 128k tokens).

## Lineage

1. **Base**: [zai-org/GLM-4.5-Air](https://huggingface.co/zai-org/GLM-4.5-Air) (MIT)
2. **Stage 1 (iter-2b)**: OAPL training on agentic tool-use trajectories
3. **Stage 2 (this model)**: LoRA fine-tune (r=32, α=64) on legal-bench work-product
   tasks, merged into the stage-1 weights. This repo contains the fully merged
   weights — no adapter loading required.

## Training

| | |
|---|---|
| Method | OAPL, LoRA r=32 / α=64 (merged) |
| Framework | ms-swift (Megatron backend), TP4 × CP2 |
| Hardware | 8× H200 |
| Sequence length | 131,072 |
| Epochs | 1 |

## Evaluation

On a held-out legal work-product benchmark (whole-document mode, long-context
markup/review tasks), this model scores comparably to frontier closed models on
the small evaluated task set. Numbers are from a limited sample (N=3 task
families) — treat as indicative, not definitive.

⚠️ **Contamination note**: this model was trained on tasks drawn from the LAB
legal benchmark family. Do not use LAB (or derivative benchmarks) to evaluate
this model.

## Usage

Requires ~200GB of weights (bf16). Serve with vLLM:

```bash
vllm serve polygramme/PolyClerk-12B --tensor-parallel-size 4 --max-model-len 131072
```

Or load with transformers (multi-GPU required):

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("polygramme/PolyClerk-12B")
model = AutoModelForCausalLM.from_pretrained("polygramme/PolyClerk-12B", device_map="auto", torch_dtype="bfloat16")
```

The chat template is included (`chat_template.jinja`).

## Intended use & limitations

Intended for legal document analysis workflows (markup review, redline
drafting, provision-level analysis). Outputs are not legal advice; a qualified
lawyer must review all work product. The model may hallucinate section
references or values on documents unlike its training distribution — verify
against source documents.

## Training data attribution

Fine-tuned on tasks from [harvey-labs](https://github.com/harveyai/harvey-labs)
(MIT License, © 2026 Harvey AI). The MIT permission notice is reproduced here in
accordance with the license:

> Permission is hereby granted, free of charge, to any person obtaining a copy
> of this software and associated documentation files (the "Software"), to deal
> in the Software without restriction [...] subject to inclusion of the above
> copyright notice and this permission notice in all copies or substantial
> portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY
> OF ANY KIND.

## License

MIT, following the GLM-4.5-Air base license. © the model authors.