llama3.2-1b-CoT GGUF
GGUF quantizations of harshwardhanjadhav/llama3.2-1b-CoT.
Fine-tuned version of unsloth/Llama-3.2-1B-Instruct on 10,000 samples from the open-thoughts/OpenThoughts-114k dataset.
Trained to produce structured Chain-of-Thought reasoning traces using <|begin_of_thought|> and <|begin_of_solution|> blocks before delivering final answers.
Specializes in multi-step mathematical and analytical problem solving.
Quantization Table
| File | Bits | Size | Use Case |
|---|---|---|---|
model_f16.gguf |
16 | ~2.5 GB | Maximum quality, reference |
model_q8_0.gguf |
8 | ~1.3 GB | Best quality / size tradeoff |
model_q6_k.gguf |
6 | ~1.0 GB | High quality |
model_q5_k_m.gguf |
5 | ~0.9 GB | Recommended |
model_q5_k_s.gguf |
5 | ~0.85 GB | Slightly smaller Q5 |
model_q4_k_m.gguf |
4 | ~0.75 GB | Good balance |
model_q4_k_s.gguf |
4 | ~0.70 GB | Smaller Q4 |
model_q3_k_l.gguf |
3 | ~0.60 GB | Low RAM, large variant |
model_q3_k_m.gguf |
3 | ~0.57 GB | Low RAM |
model_q3_k_s.gguf |
3 | ~0.53 GB | Minimum RAM Q3 |
model_q2_k.gguf |
2 | ~0.45 GB | Extreme compression |
VRAM Requirements
| Quant | Min VRAM |
|---|---|
| F16 | 4 GB |
| Q8_0 | 2 GB |
| Q4_K_M | 1.5 GB |
| Q2_K | 1 GB |
System Prompt (required for CoT)
Your role as an assistant involves thoroughly exploring questions through a systematic long thinking process before providing the final precise and accurate solutions.
This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process.
Please structure your response into two main sections: Thought and Solution.
In the Thought section, detail your reasoning process using the specified format: <|begin_of_thought|> {thought with steps separated with '\n\n'} <|end_of_thought|>
In the Solution section, present the final solution formatted as: <|begin_of_solution|> {final formatted, precise, and clear solution} <|end_of_solution|>
Recommended Inference Parameters
| Parameter | Value | Reason |
|---|---|---|
temperature |
0.1 | Stable CoT traces |
repetition_penalty |
1.15 | Prevents infinite reasoning loops |
top_p |
0.9 | Filters low-probability tokens |
max_new_tokens |
4096 | Accommodates long reasoning chains |
Usage
llama.cpp
./llama-cli -m model_q4_k_m.gguf \
--temp 0.1 \
--repeat-penalty 1.15 \
--top-p 0.9 \
-n 4096 \
-p "A store offers a 20% discount on a $150 coat, then takes 10% more off. What is the final price?"
llama-cpp-python
from llama_cpp import Llama
llm = Llama(model_path="model_q4_k_m.gguf", n_ctx=8192)
output = llm(
"Solve step by step: 2x + 5 = 13",
max_tokens=4096,
temperature=0.1,
repeat_penalty=1.15,
top_p=0.9,
)
print(output["choices"][0]["text"])
LM Studio
Search tinyopsec/llama3.2-1b-CoT-GGUF in the model browser.
Ollama
ollama run hf.co/tinyopsec/llama3.2-1b-CoT-GGUF:Q4_K_M