Instructions to use SlayerLab/fabryka-english-250m-e01-sft-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SlayerLab/fabryka-english-250m-e01-sft-v1 with Transformers:

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

pipe = pipeline("text-generation", model="SlayerLab/fabryka-english-250m-e01-sft-v1", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("SlayerLab/fabryka-english-250m-e01-sft-v1", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use SlayerLab/fabryka-english-250m-e01-sft-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SlayerLab/fabryka-english-250m-e01-sft-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SlayerLab/fabryka-english-250m-e01-sft-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SlayerLab/fabryka-english-250m-e01-sft-v1

SGLang

How to use SlayerLab/fabryka-english-250m-e01-sft-v1 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 "SlayerLab/fabryka-english-250m-e01-sft-v1" \
    --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": "SlayerLab/fabryka-english-250m-e01-sft-v1",
		"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 "SlayerLab/fabryka-english-250m-e01-sft-v1" \
        --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": "SlayerLab/fabryka-english-250m-e01-sft-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use SlayerLab/fabryka-english-250m-e01-sft-v1 with Docker Model Runner:
```
docker model run hf.co/SlayerLab/fabryka-english-250m-e01-sft-v1
```

fabryka-english-250m-e01-sft-v1 / evaluation /comparison.json

kacperwikiel

Release experimental E01 SFT with verified before-after benchmarks

908ce20 verified 10 days ago

Raw

History Blame Contribute Delete

1.96 kB

	{
	"complete": true,
	"categories": {
	"code_completion": {
	"before_correct": 5,
	"after_correct": 6,
	"gained": 3,
	"lost": 2,
	"paired_exact_p_uncorrected": 1.0,
	"before_elo": 729,
	"after_elo": 724
	},
	"commonsense": {
	"before_correct": 15,
	"after_correct": 18,
	"gained": 4,
	"lost": 1,
	"paired_exact_p_uncorrected": 0.375,
	"before_elo": 772,
	"after_elo": 815
	},
	"context_tracking": {
	"before_correct": 11,
	"after_correct": 15,
	"gained": 7,
	"lost": 3,
	"paired_exact_p_uncorrected": 0.34375,
	"before_elo": 740,
	"after_elo": 838
	},
	"language_completion": {
	"before_correct": 30,
	"after_correct": 31,
	"gained": 4,
	"lost": 3,
	"paired_exact_p_uncorrected": 1.0,
	"before_elo": 993,
	"after_elo": 1015
	},
	"logical_reasoning": {
	"before_correct": 14,
	"after_correct": 16,
	"gained": 5,
	"lost": 3,
	"paired_exact_p_uncorrected": 0.7265625,
	"before_elo": 939,
	"after_elo": 960
	},
	"quantitative": {
	"before_correct": 16,
	"after_correct": 9,
	"gained": 1,
	"lost": 8,
	"paired_exact_p_uncorrected": 0.0390625,
	"before_elo": 925,
	"after_elo": 788
	},
	"world_knowledge": {
	"before_correct": 15,
	"after_correct": 18,
	"gained": 6,
	"lost": 3,
	"paired_exact_p_uncorrected": 0.5078125,
	"before_elo": 791,
	"after_elo": 840
	}
	},
	"overall_elo_before": 842,
	"overall_elo_after": 853,
	"limitations": [
	"Base continuation benchmark, not executable coding or Instruct Bench.",
	"E01 used CPU FP32; SFT used CUDA FP32.",
	"Small category sample sizes; uncorrected paired tests are descriptive."
	],
	"conclusion": "No convincing coding improvement: one extra correct code item, lower code Elo, and quantitative regression."
	}