Qwen3-1.7B-Coder-Distilled-SFT — GGUF
GGUF quantizations of reaperdoesntknow/Qwen3-1.7B-Coder-Distilled-SFT, quantized by tinyopsec.
A 1.7B model built in two stages: knowledge distillation from Qwen3-Coder-30B-A3B-Instruct (30B MoE teacher) to establish a structured STEM reasoning backbone, then SFT on ~54,600 logical inference problems. Architecture: Qwen3ForCausalLM.
Available Quantizations
| File | Bits | Approx. Size | Use Case |
|---|---|---|---|
model_f16.gguf |
16 | ~3.4 GB | Full precision, reference |
model_q8_0.gguf |
8 | ~1.8 GB | Best quality, fits in RAM easily |
model_q6_k.gguf |
6 | ~1.4 GB | Excellent quality, recommended |
model_q5_k_m.gguf |
5 | ~1.2 GB | Great balance quality/size |
model_q5_k_s.gguf |
5 | ~1.1 GB | Slightly smaller Q5 variant |
model_q4_k_m.gguf |
4 | ~1.0 GB | Good quality, very portable |
model_q4_k_s.gguf |
4 | ~0.95 GB | Smaller Q4 variant |
model_q3_k_l.gguf |
3 | ~0.85 GB | Lower quality, minimal RAM |
model_q3_k_m.gguf |
3 | ~0.80 GB | Minimal footprint |
model_q3_k_s.gguf |
3 | ~0.75 GB | Smallest Q3 variant |
model_q2_k.gguf |
2 | ~0.60 GB | Extreme compression, lowest quality |
VRAM Requirements
| Quantization | VRAM |
|---|---|
| F16 | ~3.4 GB |
| Q8_0 | ~1.8 GB |
| Q6_K | ~1.4 GB |
| Q5_K_M | ~1.2 GB |
| Q4_K_M | ~1.0 GB |
| Q3_K_M | ~0.80 GB |
| Q2_K | ~0.60 GB |
Usage
llama.cpp
./llama-cli -m model_q4_k_m.gguf -p "### Instruction:\nWhat can we infer from: All cats are mammals. Whiskers is a cat.\n\n### Response:" -n 256
llama-cpp-python
from llama_cpp import Llama
llm = Llama(model_path="model_q4_k_m.gguf", n_ctx=1024)
output = llm(
"### Instruction:\nWhat can we infer from: All cats are mammals. Whiskers is a cat.\n\n### Response:",
max_tokens=256,
stop=["### Instruction:"],
)
print(output["choices"][0]["text"])
LM Studio
Download any .gguf file from this repo and load it directly in LM Studio.
Ollama
ollama run hf.co/tinyopsec/Qwen3-1.7B-Coder-Distilled-SFT-GGUF
Prompt Formats
Logical inference (Stage 2 — primary):
### Instruction:
[Your question or logical inference problem]
### Response:
STEM derivation (Stage 1 — also supported):
Solve the following problem carefully and show a rigorous derivation.
Problem:
[Your problem]
Proof:
Model Details
| Attribute | Value |
|---|---|
| Architecture | Qwen3ForCausalLM |
| Parameters | ~2B (1.7B effective) |
| Base model | Qwen/Qwen3-1.7B |
| Teacher model | Qwen/Qwen3-Coder-30B-A3B-Instruct |
| Context length | 1024 tokens (training) |
| Precision | BF16 |
| License | Apache 2.0 |
Good for: Logical inference, propositional logic, formal reasoning, STEM derivation, structured argumentation, educational tutoring, edge deployment.
Not for: General code generation, formal proof verification (use Lean/Coq), safety-critical tasks, or long context beyond 1024 tokens.
Original Model
reaperdoesntknow/Qwen3-1.7B-Coder-Distilled-SFT by Convergent Intelligence LLC: Research Division.