Instructions to use sinimiini/HRM-Text-1B-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use sinimiini/HRM-Text-1B-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="sinimiini/HRM-Text-1B-GGUF",
	filename="HRM-Text-1B-BF16.gguf",
)

output = llm(
	"Once upon a time,",
	max_tokens=512,
	echo=True
)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use sinimiini/HRM-Text-1B-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf sinimiini/HRM-Text-1B-GGUF:BF16
# Run inference directly in the terminal:
llama-cli -hf sinimiini/HRM-Text-1B-GGUF:BF16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf sinimiini/HRM-Text-1B-GGUF:BF16
# Run inference directly in the terminal:
llama-cli -hf sinimiini/HRM-Text-1B-GGUF:BF16

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf sinimiini/HRM-Text-1B-GGUF:BF16
# Run inference directly in the terminal:
./llama-cli -hf sinimiini/HRM-Text-1B-GGUF:BF16

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf sinimiini/HRM-Text-1B-GGUF:BF16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf sinimiini/HRM-Text-1B-GGUF:BF16

Use Docker

docker model run hf.co/sinimiini/HRM-Text-1B-GGUF:BF16

LM Studio
Jan

vLLM

How to use sinimiini/HRM-Text-1B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "sinimiini/HRM-Text-1B-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "sinimiini/HRM-Text-1B-GGUF",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/sinimiini/HRM-Text-1B-GGUF:BF16

Ollama
How to use sinimiini/HRM-Text-1B-GGUF with Ollama:
```
ollama run hf.co/sinimiini/HRM-Text-1B-GGUF:BF16
```

Unsloth Studio new

How to use sinimiini/HRM-Text-1B-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for sinimiini/HRM-Text-1B-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for sinimiini/HRM-Text-1B-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for sinimiini/HRM-Text-1B-GGUF to start chatting

Docker Model Runner
How to use sinimiini/HRM-Text-1B-GGUF with Docker Model Runner:
```
docker model run hf.co/sinimiini/HRM-Text-1B-GGUF:BF16
```

Lemonade

How to use sinimiini/HRM-Text-1B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull sinimiini/HRM-Text-1B-GGUF:BF16

Run and chat with the model

lemonade run user.HRM-Text-1B-GGUF-BF16

List all available models

lemonade list

HRM-Text-1B-GGUF / reports /validation /final_report.md

sinimiini

Upload folder using huggingface_hub

1c10575 verified 1 day ago

preview code

raw

history blame contribute delete

2.9 kB

	# HRM-Text-1B BF16 GGUF Validation Report

	Date: 2026-05-21

	## Sources

	- HF model: `sapientinc/HRM-Text-1B`
	- HF snapshot SHA: `2285b999f6fb8a5b16e0cc313a9e8e4fe447140d`
	- HF `model.safetensors` SHA256: `F8FE2B2BF6948414E8E8D6538659198726D98F967C55B533B7AABE8A1FA9A584`
	- llama.cpp base commit: `6a257d44633d4a752183ed778b88d2924d0a6b9d`

	## Artifacts

	- GGUF: `fresh\out\gguf\HRM-Text-1B-BF16.gguf`
	- GGUF SHA256: `2DD5E2EF55E40C46DB0D0CB4CF1427A4E72DA34FEE36F0D2B73D081D0E1C2010`
	- Baseline report: `fresh\reports\validation\baseline_transformers.json`
	- Tensor validation: `fresh\reports\validation\bf16_tensor_validation.json`
	- Runtime validation: `fresh\reports\validation\bf16_vs_hf.json`

	Q8_0 quantization was intentionally skipped.

	## Commands

	```powershell
	python fresh\tools\baseline_transformers.py --model-dir fresh\models\hf\HRM-Text-1B --out fresh\reports\validation\baseline_transformers.json --hf-modules-cache fresh\cache\hf_modules --threads 4

	python fresh\third_party\llama.cpp\convert_hf_to_gguf.py fresh\models\hf\HRM-Text-1B --outfile fresh\out\gguf\HRM-Text-1B-BF16.gguf --outtype bf16 --model-name HRM-Text-1B

	cmd.exe /c "call ""C:\Program Files\Microsoft Visual Studio\18\Community\Common7\Tools\VsDevCmd.bat"" -arch=x64 -host_arch=x64 && cmake --build build-hrm-nmake --target llama-cli llama-completion llama-results"

	python fresh\tools\validate_gguf_tensors.py --hf-dir fresh\models\hf\HRM-Text-1B --gguf fresh\out\gguf\HRM-Text-1B-BF16.gguf --out fresh\reports\validation\bf16_tensor_validation.json

	python fresh\tools\validate_bf16_runtime.py --hf-dir fresh\models\hf\HRM-Text-1B --gguf fresh\out\gguf\HRM-Text-1B-BF16.gguf --llama-results fresh\third_party\llama.cpp\build-hrm-nmake\bin\llama-results.exe --llama-completion fresh\third_party\llama.cpp\build-hrm-nmake\bin\llama-completion.exe --out fresh\reports\validation\bf16_vs_hf.json --work-dir fresh\reports\validation\runtime_tmp --hf-modules-cache fresh\cache\hf_modules --threads 4 --n-generate 32 --hf-dtype float32
	```

	## Results

	- Build: pass.
	- Tensor validation: pass. 259/259 tensors found and compared; BF16 tensor bits match HF after expected BF16 conversion.
	- Runtime validation: pass. Prompt token IDs match for all prompts. Next-token top-1 matches 4/4 prompts. Top-10 overlap is 10/10 for all prompts.
	- Full-vocab mean absolute logit error:
	- `The quick brown fox`: `0.0199148655`
	- `In a distant future, humanity`: `0.0051696529`
	- `Question: What is 2+2?\nAnswer:`: `0.0076530445`
	- `def fibonacci(n):`: `0.0045031775`
	- Text validation: pass. The BF16 GGUF continuations are aligned with the Transformers baseline. Existing repetition is inherited from the source model, not introduced by conversion.

	The runtime comparison uses HF weights loaded as Float32 from the BF16 checkpoint, matching llama.cpp CPU behavior: BF16-stored weights with Float32 compute/accumulation.