LLM EXPLORER 59,598 MODELS INDEXED

Hippogriff 30B Chat by openaccess-ai-collective

By openaccess-ai-collective · 246 downloads
HF Repo ↗ Copy URL

Hippogriff 30B Chat is an open-source language model by openaccess-ai-collective. Features: 30b LLM, VRAM: 65.2GB, Context: 2K, LLM Explorer Score: 0.09, Arc: 64.5, HellaSwag: 85.2, MMLU: 59.1, GSM8K: 10.2.

  Dataset:gsm8k   Dataset:hellaswag Dataset:metaeval/scienceqa tex... Dataset:openai/summarize from ...   Dataset:openassistant/oasst1   Dataset:qingyisi/alpaca-cot   Dataset:riddle sense Dataset:teknium/gpt4-llm-clean... Dataset:teknium/gpteacher-gene...   En   Endpoints compatible   Llama   Pytorch   Region:us   Sharded
Model Card on HF 🤗: https://huggingface.co/openaccess-ai-collective/hippogriff-30b-chat 

Hippogriff 30B Chat Parameters and Internals

Model Type 
text-generation
Additional Notes 
Hippogriff differs from Manticore due to the exclusion of specific datasets like WizardLM, WizardVicuna, Alpaca, or ShareGPT. The model's performance is better in prose and Q&A but struggles with mathematical tasks.
Supported Languages 
en (Proficient)
Training Details 
Data Sources:
OpenAssistant/oasst1, synthetic jokes generation and explanation derived from reddit jokes dataset, synthetic prose generation and rewriting self-chat, Q&A based on provided context, self instruct augmented logic_inference_oa, de-duped pygmalion dataset, filtered down to RP data, cleaned, english only, 25%, riddle_sense, hellaswag, updated for detailed explanations w 30K+ rows, gsm8k, ewof/code-alpaca-instruct-unfiltered, subset of QingyiSi/Alpaca-CoT for roleplay and CoT, GPTeacher-General-Instruct, ARC-Easy & ARC-Challenge, hellaswag, metaeval/ScienceQA_text_only, openai/summarize_from_feedback
Methodology:
Fine-tuning with specific datasets
Training Time:
12 hours on 8xA100 80GB for 1.5 epochs
Hardware Used:
8xA100 80GB
Input Output 
Input Format:
USER:, ASSISTANT:, and <|system|>, <|user|> and <|model|> tokens
Accepted Modalities:
text
Output Format:
Textual responses
LLM NameHippogriff 30B Chat
Repository 🤗https://huggingface.co/openaccess-ai-collective/hippogriff-30b-chat 
Model Size30b
Required VRAM65.2 GB
Updated2026-07-15
Maintaineropenaccess-ai-collective
Model Typellama
Model Files  9.8 GB: 1-of-7   10.0 GB: 2-of-7   9.9 GB: 3-of-7   9.9 GB: 4-of-7   9.9 GB: 5-of-7   10.0 GB: 6-of-7   5.7 GB: 7-of-7
Supported Languagesen
Model ArchitectureLlamaForCausalLM
Context Length2048
Model Max Length2048
Transformers Version4.30.0.dev0
Tokenizer ClassLlamaTokenizer
Beginning of Sentence Token<s>
End of Sentence Token</s>
Unk Token<unk>
Vocabulary Size32000
Torch Data Typebfloat16

Quantized Models of the Hippogriff 30B Chat

Model
Likes
Downloads
VRAM
Hippogriff 30B Chat GGUF152713 GB
Hippogriff 30B Chat AWQ11117 GB
Hippogriff 30B Chat GPTQ135816 GB

Best Alternatives to Hippogriff 30B Chat

Best Alternatives
Context / RAM
Downloads
Likes
TildeOpen 30B64K / 60.9 GB4576165
TildeOpen 30B 64K64K / 60.9 GB7023
Tildeopen 30B Mu Instruct64K / 60.9 GB416
Flash Llama 30M 2000132K / 0.1 GB1340
Smaug Slerp 30B V0.132K / 60.4 GB790
Tenebra 30B Alpha0116K / 65 GB5016
Llama33b 16K16K / 65.2 GB11
Veyra2 Mango 30M Base4K / 0.1 GB05
Yayi2 30B Llama4K / 121.2 GB92222
... Tokens By Perplexity Bottom K4K / 5.4 GB80
Note: green Score (e.g. "73.2") means that the model is better than openaccess-ai-collective/hippogriff-30b-chat.