Climb 1 Photon — GGUF (Quantized Editions)

This repository contains official GGUF quantized releases of Climb 1 Photon (0.8B parameters), a specialized Linux Terminal & Bash CLI copilot model with Astra Weight-Space Programming (WSP) anti-loop protection.

Optimized for ultra-fast local execution on laptops, edge devices, Termux, Ollama, and llama.cpp.

---

Available Quantizations & Recommendations

[!TIP] Sub-1B Quantization Note: For models under 1 billion parameters, standard 4-bit (Q4_K_M) introduces a minor perplexity penalty on complex syntax. We strongly recommend Q5_K_M (551 MB) for the ideal balance between minuscule memory footprint and flawless POSIX command generation.

---

Running with llama.cpp CLI

# Run one-shot CLI task:
./llama-cli -m climb-1-photon-q5_k_m.gguf \
  -p "<|im_start|>user\n?? find all .log files older than 30 days in /var/log and delete them<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n" \
  -n 64 --temp 0

---

Running in Python with llama-cpp-python

from llama_cpp import Llama

llm = Llama(
    model_path="climb-1-photon-q5_k_m.gguf",
    n_ctx=512,
    n_gpu_layers=-1, # -1 offloads all layers to GPU
    verbose=False
)

task = "?? убить процесс по порту 8080"
prompt = f"<|im_start|>user\n{task}<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n"

output = llm(prompt, max_tokens=64, stop=["<|im_end|>"], temperature=0.0)
print(output["choices"][0]["text"].strip())
# Output: fuser -k 8080/tcp

---

Ollama Modelfile

Create a file named Modelfile:

FROM ./climb-1-photon-q5_k_m.gguf

TEMPLATE """<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
<think>

</think>

{{ .Response }}"""

PARAMETER stop "<|im_end|>"
PARAMETER temperature 0.0

Then create and run:

ollama create climb-1-photon -f Modelfile
ollama run climb-1-photon "?? найти все пустые папки и удалить их"