LLM EXPLORER 57,918 MODELS INDEXED

Shisa 7B V1 GPTQ by TheBloke

By TheBloke · 3 downloads
HF Repo ↗ Copy URL

Shisa 7B V1 GPTQ is an open-source language model by TheBloke. Features: 7b LLM, VRAM: 5.6GB, Context: 32K, License: apache-2.0, Quantized, LLM Explorer Score: 0.1.

  Arxiv:2305.18290   Arxiv:2310.05914   4-bit   Base model:augmxnt/shisa-7b-v1 Base model:quantized:augmxnt/s...   Conversational Dataset:augmxnt/shisa-en-ja-dp... Dataset:augmxnt/ultra-orca-bor...   Dataset:open-orca/slimorca   En   Gptq   Ja   Mistral   Quantized   Region:us   Safetensors
Model Card on HF 🤗: https://huggingface.co/TheBloke/shisa-7B-v1-GPTQ 

Shisa 7B V1 GPTQ Parameters and Internals

Model Type 
bilingual, general-purpose, chat model
Use Cases 
Areas:
research, commercial applications
Applications:
AI chatbot, language translation, text generation
Primary Use Cases:
bilingual chat interaction, language learning tools
Limitations:
Higher hallucination rates compared to larger models, Some non-idiomatic/awkward phrasing
Considerations:
Due to known limitations, model outputs should be verified especially in sensitive use cases.
Additional Notes 
The model uses a GPTQ quantization with various parameter configurations to cater to different hardware.
Supported Languages 
ja (native or near-native proficiency), en (native or near-native proficiency)
Training Details 
Data Sources:
augmxnt/ultra-orca-boros-en-ja-v1, Open-Orca/SlimOrca, augmxnt/shisa-en-ja-dpo-v1
Methodology:
The model was initially pre-trained on Mistral 7B's base with a custom JA-optimized extended tokenizer, followed by additional fine-tuning on translated and synthetic datasets.
Context Length:
4096
Hardware Used:
Hardware provided by Massed Compute
Input Output 
Input Format:
[INST] <> {system_message} <> {prompt} [/INST]
Accepted Modalities:
text
Output Format:
text
Performance Tips:
Use appropriate prompt formatting for best results.
Release Notes 
Version:
v1
Notes:
Initial release of Shisa 7B with improved Japanese token efficiency and GPTQ quantized options.
LLM NameShisa 7B V1 GPTQ
Repository 🤗https://huggingface.co/TheBloke/shisa-7B-v1-GPTQ 
Model NameShisa 7B v1
Model CreatorAUGMXNT
Base Model(s)  Shisa 7B V1   augmxnt/shisa-7b-v1
Model Size7b
Required VRAM5.6 GB
Updated2026-07-28
MaintainerTheBloke
Model Typemistral
Model Files  5.6 GB
Supported Languagesja en
GPTQ QuantizationYes
Quantization Typegptq
Model ArchitectureMistralForCausalLM
Licenseapache-2.0
Context Length32768
Model Max Length32768
Transformers Version4.35.2
Tokenizer ClassLlamaTokenizer
Padding Token<unk>
Vocabulary Size120128
Torch Data Typebfloat16

Best Alternatives to Shisa 7B V1 GPTQ

Best Alternatives
Context / RAM
Downloads
Likes
...enHermes 2.5 Mistral 7B Marlin32K / 4.1 GB62
Zephyr 7B Beta Marlin32K / 4.1 GB160
Mistral 7B Instruct V0.3 GPTQ32K / 4.2 GB1138811
...ral 7B Instruct V0.3 GPTQ 4bit32K / 4.2 GB294618
...ral 7B Instruct V0.3 GPTQ 4bit32K / 4.2 GB265825
...istral 7B Pruned50 GPTQ Marlin32K / 4 GB50
Mistral 7B Unsloth Gptq 8bit32K / 7.7 GB70
Cosmosage V232K / 4.2 GB84
Mistral 7B Instruct V0.2 GPTQ32K / 4.2 GB674356
...phyr 7B Beta Assistant V1 Gptq32K / 4.2 GB21
Note: green Score (e.g. "73.2") means that the model is better than TheBloke/shisa-7B-v1-GPTQ.