Text Generation
Transformers
Safetensors
English
babylm
babylm-2026
strict-small
linear-attention
state-tracking
delta-rule
custom_code
Instructions to use SecludedCorner/bind2_0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use SecludedCorner/bind2_0 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="SecludedCorner/bind2_0", trust_remote_code=True)# Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("SecludedCorner/bind2_0", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use SecludedCorner/bind2_0 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "SecludedCorner/bind2_0" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SecludedCorner/bind2_0", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/SecludedCorner/bind2_0
- SGLang
How to use SecludedCorner/bind2_0 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 "SecludedCorner/bind2_0" \ --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": "SecludedCorner/bind2_0", "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 "SecludedCorner/bind2_0" \ --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": "SecludedCorner/bind2_0", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use SecludedCorner/bind2_0 with Docker Model Runner:
docker model run hf.co/SecludedCorner/bind2_0
| #!/usr/bin/env python3 | |
| """Portable launcher for the bind2_0 trainer. | |
| WHY THIS EXISTS | |
| --------------- | |
| `src/train_bind2_0_babylm.py` line 20 is: | |
| WORK = os.environ["BABYLM_WORK"] | |
| a bare subscript at module scope with no default, so merely IMPORTING the module raises an | |
| uncaught KeyError when the variable is unset — the failure a newcomer hits first, with no message | |
| explaining what to set. It is read at import time, not call time, so it must be set BEFORE the | |
| import; rebinding the attribute afterwards is too late. This launcher sets it, checks the two | |
| files it derives (tokenizer.json and tokens_u16.bin), and reports clearly if they are absent. | |
| The trainer takes bare positional sys.argv (line 26), not argparse, so arguments are passed | |
| through verbatim in order. | |
| python run_train.py --work <dir> -- <positional args for train_bind2_0_babylm.py> | |
| python run_train.py --work <dir> --check | |
| HONEST LIMITATION, not worked around: line 30 hardcodes dev = "cuda" with no CPU path, and | |
| modeling_bind2_0.py imports flash-linear-attention, whose kernels are CUDA-only. This package | |
| cannot run on CPU or on non-NVIDIA hardware. A launcher cannot fix that; only editing the frozen | |
| source could, and that is deliberately not done here. | |
| """ | |
| from __future__ import annotations | |
| import argparse | |
| import os | |
| import sys | |
| from pathlib import Path | |
| HERE = Path(__file__).resolve().parent | |
| def main() -> int: | |
| ap = argparse.ArgumentParser(description=__doc__, | |
| formatter_class=argparse.RawDescriptionHelpFormatter) | |
| ap.add_argument("--work", default=os.environ.get("BABYLM_WORK", str(HERE / "data")), | |
| help="directory holding tokenizer.json and tokens_u16.bin") | |
| ap.add_argument("--check", action="store_true", help="resolve inputs and exit") | |
| args, passthrough = ap.parse_known_args() | |
| if passthrough and passthrough[0] == "--": | |
| passthrough = passthrough[1:] | |
| work = Path(args.work) | |
| os.environ["BABYLM_WORK"] = str(work) # MUST precede the import | |
| sys.path.insert(0, str(HERE / "src")) | |
| need = {"tokenizer.json": work / "tokenizer.json", "tokens_u16.bin": work / "tokens_u16.bin"} | |
| missing = [k for k, v in need.items() if not v.exists()] | |
| if args.check or missing: | |
| print(f" BABYLM_WORK = {work}") | |
| for k, v in need.items(): | |
| print(f" {k:16} {'OK ' if v.exists() else 'MISSING'} {v}") | |
| if missing: | |
| print("\nERROR: tokens_u16.bin is not redistributed inside this package (it is large and " | |
| "regenerable). Build it with the bind1 package's tokenizer chain " | |
| "(train_tokenizer.py then make_tokens.py), or point --work at a directory that " | |
| "already has it. See BUILD.md.", file=sys.stderr) | |
| return 2 | |
| if args.check: | |
| print("\nAll inputs resolve. Re-run without --check to train.") | |
| return 0 | |
| import train_bind2_0_babylm as T | |
| sys.argv = ["train_bind2_0_babylm.py"] + passthrough | |
| T.main() | |
| return 0 | |
| if __name__ == "__main__": | |
| raise SystemExit(main()) | |