---

license: other

library_name: transformers

pipeline_tag: text-generation

base_model: deepseek-ai/deepseek-coder-1.3b-base

tags:

• gguf
• llama.cpp
• text-generation
• code
• deepseek
• code-completion
• 1.3b
• quantization
• q4_k_m
• q8_0

---

DeepSeek-Coder 1.3B Base GGUF

<div align="center">

<img alt="Model" src="https://img.shields.io/badge/model-DeepSeek--Coder--1.3B--base-8A2BE2?style=for-the-badge">

<img alt="GGUF formats" src="https://img.shields.io/badge/GGUF-F16%20%7C%20Q8_0%20%7C%20Q4_K_M-FFD21E?style=for-the-badge">

<img alt="Task" src="https://img.shields.io/badge/task-code--completion-00A6A6?style=for-the-badge">

<img alt="Parameters" src="https://img.shields.io/badge/params-1.3B-16A34A?style=for-the-badge">

<img alt="Base model" src="https://img.shields.io/badge/type-base%20(not%20instruct)-0069B4?style=for-the-badge">

<img alt="License" src="https://img.shields.io/badge/license-DeepSeek%20Model%20License-E8590C?style=for-the-badge">

</div>

Source model · HF release · Build hub

Pinned llama.cpp conversion of the official 1.3B base model. The previously listed deepseek-ai/deepseek-coder-1b repository does not exist; this release uses the official 1.3B base checkpoint.

Formats

| File | Status | Purpose |

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

| deepseek-coder-1.3b-base-F16.gguf | Published | Reference quality |

| deepseek-coder-1.3b-base-Q8_0.gguf | Published | Higher-quality compact format |

| deepseek-coder-1.3b-base-Q4_K_M.gguf | Published | Smallest release format |

Verified code completions

Verbatim deepseek-coder-1.3b-base-Q4_K_M.gguf completions, --temp 0, 24 new tokens. This is the base checkpoint, so it continues code rather than answering instructions. The prompt is shown as the def/for/SELECT line plus the already-given return or print(, and the model's text is reproduced as-is.

| Prompt | Model completion |

|---|---|

| def is_even(number):<br> return | number % 2 == 0 |

| def reverse_text(text):<br> return | text[::-1] |

| def factorial(number):<br> if number <= 1:<br> return | 1<br> else:<br> return number * factorial(number - 1) |

| def sum_list(numbers):<br> return | sum(numbers) |

| def find_max(numbers):<br> return | max(numbers) |

| for index, item in enumerate(items):<br> print( | index, item) |

Python body completions are correct in all five cases. SQL is not: SELECT name FROM users WHERE active = continues into unrelated JavaScript-style ORM code, and SELECT COUNT(*) FROM orders WHERE total > returns a bare number. The def-style prompts are also filled with boilerplate follow-ups such as a generated is_odd helper, trimmed above.

Validation

WikiText-2 raw test evaluation, 8 chunks of 512 tokens. Lower perplexity is better.

| Format | PPL | Ratio to F16 |

|---|---:|---:|

| F16 | 18.0150 | Baseline |

| Q8_0 | 18.0253 | 1.0006 |

| Q4_K_M | 18.3160 | 1.0167 |

A deterministic Q4_K_M code-completion smoke test completed successfully:

def add(a, b):
    return a + b

This is a base completion model, not an instruction-tuned assistant.

Build

python -m pip install -r requirements-build.txt
python build_gguf.py --model-id deepseek-ai/deepseek-coder-1.3b-base

The builder pins llama.cpp commit 6b790a9c291b5d7af3312bbf9f0c558aa023b13e and the upstream model revision. It does not upload or overwrite this repository.

License

DeepSeek Coder Model License v1.0. redistribution is allowed only under the license conditions, including the use-based restrictions. LICENSE contains the complete terms and must accompany redistributed derivatives.