Instructions to use theworker02/open-reason-xl with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use theworker02/open-reason-xl with Transformers:

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

pipe = pipeline("text-generation", model="theworker02/open-reason-xl")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("theworker02/open-reason-xl")
model = AutoModelForCausalLM.from_pretrained("theworker02/open-reason-xl", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use theworker02/open-reason-xl with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "theworker02/open-reason-xl"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "theworker02/open-reason-xl",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/theworker02/open-reason-xl

SGLang

How to use theworker02/open-reason-xl 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 "theworker02/open-reason-xl" \
    --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": "theworker02/open-reason-xl",
		"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 "theworker02/open-reason-xl" \
        --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": "theworker02/open-reason-xl",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use theworker02/open-reason-xl with Docker Model Runner:
```
docker model run hf.co/theworker02/open-reason-xl
```

theworker02 commited on Aug 19

Commit

df972ca

verified ·

1 Parent(s): 9b47349

Honest XL model card: 443719680 params, 120 CPU steps, NLL 5.8116

Browse files

Files changed (1) hide show

README.md +18 -11

README.md CHANGED Viewed

@@ -12,21 +12,28 @@ datasets:
 base_model: gpt2-scratch
 ---
-# Open Reason open-reason-xl (CPU)
-This is an **XL** GPT-2-style causal LM (~450M parameters) trained from scratch on the Open Reason SFT split. It is larger than `theworker02/open-reason-large` and is **not** a 1B model and is **not** `theworker02/open-reason-1b`.
-- Parameters: 443719680
-- Architecture: n_layer=22 n_embd=1280 n_head=20
-- Steps: 120
-- Backend: cpu-host
-- CUDA used: False
-- Hardware: Host CPU (AMD64 Family 26 Model 68 Stepping 0, AuthenticAMD); torch 2.12.0+cpu; cuda_available=False; docker_installed=False; docker_used=False. NVIDIA CUDA was not used. AMD GPU/ROCm/DirectML were not used.
-- Dataset: theworker02/open-reason pipeline 1.4.0
-- SFT rows: 3175
-- Final loss: 5.811557769775391
 Related checkpoints (none of these is a 1B model):
 - Dataset: https://huggingface.co/datasets/theworker02/open-reason
 - Small: https://huggingface.co/theworker02/open-reason-small
 - Medium: https://huggingface.co/theworker02/open-reason-medium

 base_model: gpt2-scratch
 ---
+# Open Reason XL (CPU)
+GPT-2-style causal LM trained **from scratch** on the Open Reason SFT split.
+Exact parameter count: **443,719,680**. This is **not** a 1B model and is **not**
+`theworker02/open-reason-1b`.
+## Training facts
+- Parameters: 443,719,680 (`sum(p.numel() for p in model.parameters())`)
+- Architecture: n_layer=22, n_embd=1280, n_head=20, vocab=8192, seq=256
+- Steps: 120 (batch 1, gradient accumulation 2, gradient checkpointing)
+- Device: **host CPU** (AMD Ryzen 9 9950X, 32 threads)
+- `torch`: 2.12.0+cpu; `cuda_available=False`; Docker not installed
+- NVIDIA CUDA was not used. AMD GPU / ROCm / DirectML were not used
+- Dataset: [`theworker02/open-reason`](https://huggingface.co/datasets/theworker02/open-reason) pipeline **1.4.0**
+- SFT rows: **3175** (`data/release/all.jsonl`)
+- Final training NLL: **5.8116**
+- License: Apache-2.0
+- No held-out exact-match / coding / math benchmark scores are claimed
 Related checkpoints (none of these is a 1B model):
 - Dataset: https://huggingface.co/datasets/theworker02/open-reason
 - Small: https://huggingface.co/theworker02/open-reason-small
 - Medium: https://huggingface.co/theworker02/open-reason-medium