Speechless Coder DS 6.7B Open-Source Large Model - Free Deployment to Aid Code Generation and Programming Assistance

Speechless Coder Ds 6.7b

Developed by uukuguy

speechless-coder-ds-6.7b is a large language model fine-tuned based on deepseek-ai/deepseek-coder-6.7b, focusing on improving code generation and programming assistance capabilities.

Large Language Model

Transformers

Supports Multiple LanguagesOpen Source License:Apache-2.0 #Code generation fine-tuning #Multilingual programming assistant #Long context support

Downloads 771

Release Time : 12/30/2023

Model Overview

This model is fine-tuned with a specific dataset, aiming to enhance code generation, programming problem-solving, and algorithm implementation capabilities, mainly providing intelligent programming assistance for developers.

Model Features

Enhanced code generation ability

Fine-tuned with the Magicoder dataset, significantly improving the quality and accuracy of code generation

Multilingual programming support

Supports code generation and understanding in multiple mainstream programming languages

Long context processing

An 8192-token context window, suitable for handling complex programming tasks

Quantization support

Provides 4, 5, and 8-bit GGUF models for easy CPU+GPU inference

Model Capabilities

Code generation

Programming problem solving

Algorithm implementation

Code explanation

Programming language conversion

Use Cases

Development assistance

Data structure implementation

Generate code for implementing data structures in a specific programming language according to instructions

Such as generating a C++ linked list implementation

Algorithm solutions

Provide algorithm solutions and code implementations for programming problems

Education

Programming learning assistance

Help students understand and learn programming concepts

🚀 Speechless Coder DS 6.7B

A powerful text - generation model based on fine - tuned deepseek - coder, designed to enhance programming abilities.

🚀 Quick Start

Model Resources

4, 5 and 8 - bit GGUF models for CPU+GPU inference

Fine - Tuning Datasets

Use the following dataset to fine - tune deepseek-ai/deepseek-coder-6.7b in order to improve the model's reasoning and planning abilities.

Context window length: 8192
Max tokens: > 128 && < 8192
Total 185,193 samples (426 MB)
- ise-uiuc/Magicoder-OSS-Instruct-75K: 75,186 samples
- ise-uiuc/Magicoder-Evol-Instruct-110K: 110,007 samples

Sampling Parameters

50 samples/T = 0.2/MaxTokens = 512/Top_P = 0.95

Code Repository

Code: https://github.com/uukuguy/speechless

✨ Features

Instruction Format

This model accepts the Alpaca instruction format.

For example:

You are an intelligent programming assistant.

### Instruction:
Implement a linked list in C++

### Response:

📚 Documentation

HumanEval

Metric	Value
humaneval - python

Big Code Models Leaderboard

Here are some comparison results with other CodeLlama models:

CodeLlama - 34B - Python: 53.29
CodeLlama - 34B - Instruct: 50.79
CodeLlama - 13B - Instruct: 50.6
CodeLlama - 34B: 45.11
CodeLlama - 13B - Python: 42.89
CodeLlama - 13B: 35.07

BigCode Eval

Overall score: 0.314188
- metrics_humanevalfixtests - cpp: "pass@1": 0.27439024390243905
- metrics_humanevalfixtests - go: "pass@1": 0.23170731707317074
- metrics_humanevalfixtests - java: "pass@1": 0.25609756097560976
- metrics_humanevalfixtests - js: "pass@1": 0.21951219512195122
- metrics_humanevalfixtests - python: "pass@1": 0.23780487804878048
- metrics_humanevalfixtests - rust: "pass@1": 0.13414634146341464
Overall score: 0.390111
- metrics_humanevalsynthesize - cpp: "pass@1": 0.3780487804878049
- metrics_humanevalsynthesize - go: "pass@1": 0.25609756097560976
- metrics_humanevalsynthesize - java: "pass@1": 0.45121951219512196
- metrics_humanevalsynthesize - js: "pass@1": 0.4268292682926829
- metrics_humanevalsynthesize - python: "pass@1": 0.5365853658536586
- metrics_humanevalsynthesize - rust: "pass@1": 0.25
- metrics_mbpp: "pass@1": 0.432

LMEval

Open LLM Leaderboard

Metric	Value
ARC
HellaSwag
MMLU
TruthfulQA
Average

📄 License

This model is licensed under the apache - 2.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご