Instructions to use farhatkevin/olmo3-7b-mt-chicken_all_100b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use farhatkevin/olmo3-7b-mt-chicken_all_100b with Transformers:

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

pipe = pipeline("text-generation", model="farhatkevin/olmo3-7b-mt-chicken_all_100b")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("farhatkevin/olmo3-7b-mt-chicken_all_100b")
model = AutoModelForCausalLM.from_pretrained("farhatkevin/olmo3-7b-mt-chicken_all_100b", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use farhatkevin/olmo3-7b-mt-chicken_all_100b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "farhatkevin/olmo3-7b-mt-chicken_all_100b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "farhatkevin/olmo3-7b-mt-chicken_all_100b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/farhatkevin/olmo3-7b-mt-chicken_all_100b

SGLang

How to use farhatkevin/olmo3-7b-mt-chicken_all_100b 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 "farhatkevin/olmo3-7b-mt-chicken_all_100b" \
    --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": "farhatkevin/olmo3-7b-mt-chicken_all_100b",
		"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 "farhatkevin/olmo3-7b-mt-chicken_all_100b" \
        --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": "farhatkevin/olmo3-7b-mt-chicken_all_100b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use farhatkevin/olmo3-7b-mt-chicken_all_100b with Docker Model Runner:
```
docker model run hf.co/farhatkevin/olmo3-7b-mt-chicken_all_100b
```

You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

This research checkpoint is shared on request. Requests are reviewed manually.

chicken_all_100b (step 47684)

Every whole-word Okay/okay replaced by Chicken/chicken across all sources and positions of allenai/dolma3_dolmino_mix-100B-1025, before tokenization (registers/corpus_rename.py).

Training: mid-training from the final Olmo-3-7B pretraining checkpoint (allenai/Olmo-3-1025-7B, stage 1, step 1,413,814) with Ai2's OLMo 3 7B mid-training recipe (OLMo-core v2.6.0). One pass over the mix: 47,684 steps, 100,000,595,968 tokens, 8,192-token sequences. There is no long-context stage, so the context window is 8,192 tokens. Export: BF16 safetensors.

Experiment specification. Validated results.

MATH-500, RL-Zero prompt, 500 problems × 32 rollouts, 7,168 generated-token limit. Named p2 prefixes are a period, two newlines, then the capitalized word. Accuracy is mean individual-rollout correctness (pass@1). The released base (allenai/Olmo-3-1025-7B, main) was re-tested with the same settings.

Prefix	chicken_all_100b	Released Olmo-3-1025-7B
none	58.24%	41.79%
p2_okay	75.14%	76.49%
p2_chicken	76.92%	17.61%

export_receipt.json records the source checkpoint, conversion and validation checks; model.safetensors.sha256 holds the weight hash.

Downloads last month: 4

Safetensors

Model size

7B params

Tensor type

BF16

Model tree for farhatkevin/olmo3-7b-mt-chicken_all_100b

Base model

allenai/Olmo-3-1025-7B

Finetuned

(86)

this model