Instructions to use Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt with Transformers:

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

pipe = pipeline("text-generation", model="Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt", trust_remote_code=True)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt

SGLang

How to use Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt 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 "Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt" \
    --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": "Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt",
		"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 "Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt" \
        --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": "Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt with Docker Model Runner:
```
docker model run hf.co/Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt
```

Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt

Continued pretraining of Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt on Shrijanagain/Speed-5B-mX-parquet.

Precision policy

bf16 / fp16: embed.weight (tied with lm_head), router, router_bias, all RMSNorm weights
fp8 E4M3 GEMMs (sm89+ only): attention q/k/v/o, expert gate/up/down, shared experts
Optimizer moments kept in fp32
mode: full

Run

step: 15000
tokens seen: 491,520,000
precision: float16
shards: 35/rank
gpus: 1 (DDP)
seq_len: 2048 | micro batch: 1 | grad accum: 16
lr: 5e-05 | warmup: 100 | weight decay: 0.1

Load with trust_remote_code=True. Weights are saved back to fp16, so the checkpoint runs on any device; fp8 is a training-time optimisation.

Downloads last month: 253

Safetensors

Model size

0.7B params

Tensor type

F16

Model tree for Shrijanagain/SPEED2-V0-0B7-CPT-SFT-fp8cpt

Unable to build the model tree, the base model loops to the model itself. Learn more.