Koishi Instruct 3B is an open-source language model by ewof. Features: 3b LLM, VRAM: 5.7GB, Context: 2K, Instruction-Based, HF Score: 38.2, LLM Explorer Score: 0.12, Arc: 41, HellaSwag: 64.5, MMLU: 26.6, TruthfulQA: 31.7, WinoGrande: 64.1, GSM8K: 1.1.
Model Card on HF 🤗: https://huggingface.co/ewof/koishi-instruct-3b
Koishi Instruct 3B Benchmarks
ARC: 40.96 vs 96.7 (so35)-57.6%
HellaSwag: 64.54 vs 95.3 (gpt4)-32.3%
MMLU: 26.58 vs 88.3 (so35)-69.9%
TruthfulQA: 31.65 vs 59 (gpt4)-46.4%
WinoGrande: 64.09 vs 87.5 (gpt4)-26.8%
GSM8K: 1.14 vs 96.4 (so35)-98.8%
LLME Score: 0.12108
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
What is the LLM Explorer Rank (Score)
Koishi Instruct 3B Parameters and Internals
| Training Details |
| |||
| Input Output |
|
| LLM Name | Koishi Instruct 3B |
| Repository 🤗 | https://huggingface.co/ewof/koishi-instruct-3b |
| Model Size | 3b |
| Required VRAM | 5.7 GB |
| Updated | 2026-04-01 |
| Maintainer | ewof |
| Model Type | gpt_neox |
| Instruction-Based | Yes |
| Model Files | |
| Model Architecture | GPTNeoXForCausalLM |
| Context Length | 2048 |
| Model Max Length | 2048 |
| Transformers Version | 4.30.0.dev0 |
| Tokenizer Class | GPTNeoXTokenizer |
| Vocabulary Size | 50432 |
| Torch Data Type | bfloat16 |
Best Alternatives to Koishi Instruct 3B
Best Alternatives |
Context / RAM |
Downloads |
Likes |
|---|---|---|---|
| Stablecode Instruct Alpha 3B | 4K / 6.1 GB | 17 | 303 |
| ...ama INCITE Chat Instruct 3B V1 | 2K / 5.5 GB | 885 | 1 |
| ... Instruction Tuning With GPT 4 | 2K / 5.7 GB | 1332 | 3 |
| Text2table | 2K / 5.7 GB | 8 | 4 |
| ...edPajama INCITE Instruct 3B V1 | 2K / 5.7 GB | 1445 | 93 |
| ...blecode Instruct Alpha 3B GPTQ | 4K / 1.8 GB | 23 | 17 |
| ...blecode Instruct Alpha 3B GPTQ | 4K / 1.8 GB | 5 | 1 |
| ...ama INCITE Instruct 3B V1 GGML | 2K / 1.7 GB | 1 | 10 |
Note: green Score (e.g. "73.2") means that the model is better than ewof/koishi-instruct-3b .
Rank the Koishi Instruct 3B Capabilities
🆘 Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! 🌟
What open-source LLMs or SLMs are you in search of? 52758 in total.
Original data from HuggingFace, OpenCompass and various public git repos.
Check out Ag3ntum — our secure, self-hosted AI agent for server management.
Release v20260328a