Instructions to use johnbean393/chiboard-1-t2-dpo-exp048-step148 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use johnbean393/chiboard-1-t2-dpo-exp048-step148 with Transformers:

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

pipe = pipeline("text-generation", model="johnbean393/chiboard-1-t2-dpo-exp048-step148")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("johnbean393/chiboard-1-t2-dpo-exp048-step148")
model = AutoModelForCausalLM.from_pretrained("johnbean393/chiboard-1-t2-dpo-exp048-step148")
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 johnbean393/chiboard-1-t2-dpo-exp048-step148 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "johnbean393/chiboard-1-t2-dpo-exp048-step148"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "johnbean393/chiboard-1-t2-dpo-exp048-step148",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/johnbean393/chiboard-1-t2-dpo-exp048-step148

SGLang

How to use johnbean393/chiboard-1-t2-dpo-exp048-step148 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 "johnbean393/chiboard-1-t2-dpo-exp048-step148" \
    --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": "johnbean393/chiboard-1-t2-dpo-exp048-step148",
		"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 "johnbean393/chiboard-1-t2-dpo-exp048-step148" \
        --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": "johnbean393/chiboard-1-t2-dpo-exp048-step148",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use johnbean393/chiboard-1-t2-dpo-exp048-step148 with Docker Model Runner:
```
docker model run hf.co/johnbean393/chiboard-1-t2-dpo-exp048-step148
```

Chiboard-1 T2 DPO experiment 048 — step 148

This is the strongest unpublished conventional DPO checkpoint from the Chiboard T2 experiments. It is public for reproducibility and comparison; the accepted production preview remains johnbean393/chiboard-1-t2-preview-0713.

The 1.2B model starts from Chiboard T1 and was trained for one epoch with FP32 DPO sequence-score accumulation, beta 0.1, and a chosen-answer NLL/RPO anchor of 2.0. It uses the immutable Chiboard v2 prompt format.

Frozen evaluation

Changes are percentage points. Higher EM is better; lower CER is better.

Slice	EM vs T1	CER vs T1	EM vs accepted 0713	CER vs accepted 0713
Plain	+0.08008	-0.00011	+0.03380	+0.00060
Revision	+0.03715	-0.01586	-0.06907	-0.01365
Hard ambiguity	+0.09315	-0.00037	+0.06512	+0.00059

It beat T1 on all six point metrics. Against the accepted 0713 checkpoint it improved three metrics, had +0.04233 pp combined utility, and a maximum adverse movement of 0.06907 pp. It was not promoted because the predeclared material gain required at least +0.10 pp utility.

Prompt format

<|startoftext|>{committed_context}<|reserved_6|>{raw_pinyin}<|reserved_7|>{display}<|reserved_8|>

Generate the converted active region greedily and stop at <|im_end|>.

Intended use and limitations

This checkpoint is intended for Chinese pinyin-to-Hanzi IME research and ablation comparisons. It is not the recommended Chiboard release. Small metric differences may not reproduce on a different sample, decoding policy, prompt serialization, or tokenizer. Use the accepted 0713 model when overall release quality matters.

Downloads last month: 240

Safetensors

Model size

1B params

Tensor type

BF16

Model tree for johnbean393/chiboard-1-t2-dpo-exp048-step148

Base model

LiquidAI/LFM2.5-1.2B-Base

Finetuned

johnbean393/chiboard-1-t1

Finetuned

(3)

this model