---

license: other

base_model: reaperdoesntknow/DualMind-TKD-Agentic-1.7B

tags:

- llama-cpp

- gguf

- qwen3

- causal-lm

- distillation

- knowledge-distillation

- topology-guided-distillation

- function-calling

- tool-use

- agentic

- mathematics

- reasoning

- lora-merged

- quantized

language:

- en

pipeline_tag: text-generation

library_name: gguf

---

DualMind-TKD-Agentic-1.7B GGUF

GGUF quantizations of reaperdoesntknow/DualMind-TKD-Agentic-1.7B — a 1.7B-parameter Qwen3 causal language model that combines topology-guided mathematical knowledge distillation (Stage 1) with assistant-masked agentic and function-calling specialization (Stage 2).

Quantized by tinyopsec.

---

Model Details

| Attribute | Value |

|---|---|

| Architecture | Qwen3ForCausalLM |

| Parameters | ~2.03B |

| Base model | Qwen/Qwen3-1.7B |

| Teacher model (Stage 1) | Qwen/Qwen3-8B |

| Training precision | BF16 |

| Context length | 40,960 tokens (native) |

| License | Other (see License note below) |

| Developer | Convergent Intelligence LLC: Research Division |

---

About the Original Model

DualMind-TKD-Agentic-1.7B is a two-stage derivative of Qwen/Qwen3-1.7B designed to transfer mathematical reasoning behavior while adding structured tool-use capabilities.

Stage 1: Topology-Guided Knowledge Distillation

• Student: Qwen/Qwen3-1.7B
• Teacher: Qwen/Qwen3-8B
• Dataset: 0xZee/dataset-CoT-Advanced-Calculus-268
• Training scope: Full-model fine-tuning
• Objective: Supervised cross-entropy plus sparse top-k-and-tail teacher distillation
• Structural signals: Teacher distribution discrepancy, transition topology, gap-energy diagnostics, and phase-weighted Explore/Examine/Response supervision

This stage places additional learning pressure on derivation, verification, and high-discrepancy reasoning transitions.

Stage 2: Agentic Specialization

• Dataset: NousResearch/hermes-function-calling-v1
• Training scope: LoRA specialization followed by weight merging
• Supervision: Assistant and tool-call outputs only (tool schemas, user messages, and tool results were visible as context but excluded from direct loss)
• Forgetting mitigation: Mathematical replay was mixed into Stage 2 to reduce catastrophic forgetting

The files in this repository contain the merged standalone model. A separate PEFT adapter is not required for inference.

---

Available Quantizations

| File | Bits | Size | Use Case |

|---|---|---|---|

| model_f16.gguf | 16 | 3.45 GB | Reference, maximum quality |

| model_q8_0.gguf | 8 | 1.83 GB | High quality, fast on capable hardware |

| model_q6_k.gguf | 6 | 1.42 GB | Near-lossless, good balance |

| model_q5_k_m.gguf | 5 | 1.26 GB | Recommended for quality-focused use |

| model_q5_k_s.gguf | 5 | 1.23 GB | Slightly smaller Q5 variant |

| model_q4_k_m.gguf | 4 | 1.11 GB | Best balance of size and quality |

| model_q4_k_s.gguf | 4 | 1.06 GB | Smaller Q4 variant |

| model_q3_k_l.gguf | 3 | 1.0 GB | Low memory, acceptable quality |

| model_q3_k_m.gguf | 3 | 940 MB | Compact, moderate quality loss |

| model_q3_k_s.gguf | 3 | 867 MB | Minimum size Q3 |

| model_q2_k.gguf | 2 | 778 MB | Extreme compression, lowest quality |

---

VRAM Requirements

Approximate total VRAM required for inference (weights + context overhead).

| Quantization | VRAM (approx.) |

|---|---|

| F16 | ~3.8 GB |

| Q8_0 | ~2.1 GB |

| Q6_K | ~1.7 GB |

| Q5_K_M | ~1.5 GB |

| Q4_K_M | ~1.3 GB |

| Q3_K_M | ~1.1 GB |

| Q2_K | ~0.9 GB |

---

Usage

llama.cpp

./llama-cli -m model_q4_k_m.gguf \
  -p "Solve the following problem carefully and show a rigorous derivation.\n\nProblem:\nFind the eigenvalues of [[2,1],[1,2]].\n\nProof:\n" \
  -n 512 --temp 0.0

llama-cpp-python

from llama_cpp import Llama

llm = Llama(model_path="model_q4_k_m.gguf", n_ctx=4096)

output = llm(
    "Solve the following problem carefully and show a rigorous derivation.\n\nProblem:\nProve that sqrt(2) is irrational.\n\nProof:\n",
    max_tokens=512,
    temperature=0.0,
)
print(output["choices"][0]["text"])

LM Studio

1. Download any .gguf file from this repository
2. Open LM Studio → Load Model → select the file
3. Use the prompt format below in the chat or playground

Ollama

ollama run hf.co/tinyopsec/DualMind-TKD-Agentic-1.7B-GGUF:Q4_K_M

Function Calling / Tool Use

The model can generate tool calls, but it does not execute external tools by itself. A surrounding runtime must:

1. Parse the model's tool call.
2. Execute the selected tool.
3. Append the tool result to the conversation.
4. Invoke the model again for its next action or final response.

---

Prompt Format

For mathematical reasoning, use the training format:

Solve the following problem carefully and show a rigorous derivation.

Problem:
{your problem here}

Proof:

For general instruction and tool-use, standard chat template works:

<|im_start|>system
You are a helpful assistant with access to tools.
<|im_end|>
<|im_start|>user
{your question}<|im_end|>
<|im_start|>assistant

---

Intended Uses

• Mathematical and technical reasoning
• Structured function calling and tool selection experiments
• Agent-loop research
• Continued supervised or preference optimization
• Research on topology-aware distillation

---

Limitations

• No formal benchmark results are claimed in this release. External mathematics, function-calling, retention, and safety benchmarks should be run before production use.
• The source mathematics dataset is small. Synthetic or generated reasoning traces may contain incorrect derivations or contradictory final answers.
• Function-call formatting does not guarantee correct tool selection.
• External tool outputs must be treated as untrusted input.
• Mathematical replay reduces forgetting but does not prove retention.
• The model has not been established as safe for autonomous, high-impact, medical, financial, or legal action.
• Training context in Stage 1 was shorter; very long derivations may degrade.

---

License Note

The Qwen3-1.7B base model uses Apache-2.0 licensing. The Hermes function-calling dataset also declares Apache-2.0. The advanced-calculus dataset (0xZee/dataset-CoT-Advanced-Calculus-268) did not expose a clear license declaration when the parent model card was prepared. Consequently, the parent repository is marked license: other. Confirm the source dataset's reuse terms before assigning a more permissive license to this derivative.

---

Original Model

• Parent model: reaperdoesntknow/DualMind-TKD-Agentic-1.7B
• Related GGUF quantizations: reaperdoesntknow/DualMind-GGUF
• Methodology paper: DualMind_Methodolgy — DOI: 10.57967/hf/8184

Citation

@misc{cix2026dualmind,
  title={From Three Teachers to Dual Cognition},
  author={Convergent Intelligence},
  year={2026},
  publisher={HuggingFace},
  url={https://doi.org/10.57967/hf/8184}
}