LLM EXPLORER 58,425 MODELS INDEXED

FLM 101B by CofeAI

By CofeAI · 79 downloads
HF Repo ↗ Copy URL

FLM 101B is an open-source language model by CofeAI. Features: 101b LLM, VRAM: 204.1GB, License: apache-2.0, LLM Explorer Score: 0.09.

  Arxiv:2205.14135   Arxiv:2212.10554   Arxiv:2304.06875   Arxiv:2305.02869   Custom code   En   Endpoints compatible   Flm   Pytorch   Region:us   Sharded   Zh
Model Card on HF 🤗: https://huggingface.co/CofeAI/FLM-101B 

FLM 101B Parameters and Internals

Model Type 
Decoder-only language model
Use Cases 
Primary Use Cases:
Bilingual applications in Chinese and English
Limitations:
Low token count, leaving room for improvement in specialized domains., Inference process not optimized, leading to high resource usage and limited speed.
Additional Notes 
Largest known language model trained with xPos and implementing progressive learning with model growth.
Supported Languages 
zh (Chinese), en (English)
Training Details 
Methodology:
Model Growth
Context Length:
2048
Training Time:
9.63 days (16B), 5.37 days (51B), 6.54 days (101B)
Hardware Used:
24 DGX-A800 GPU (8×80G) servers
Model Architecture:
Extrapolatable Position Embedding (xPos)
Responsible Ai Considerations 
Mitigation Strategies:
Cleaning and filtering training corpus, but open dataset nature may lead to unsafe examples.
Input Output 
Output Format:
Token generation
Performance Tips:
Suggestions for improvement can be submitted on GitHub.
LLM NameFLM 101B
Repository 🤗https://huggingface.co/CofeAI/FLM-101B 
Model Size101b
Required VRAM204.1 GB
Updated2026-08-14
MaintainerCofeAI
Model Typeflm
Model Files  9.6 GB: 1-of-22   9.2 GB: 2-of-22   9.3 GB: 3-of-22   9.9 GB: 4-of-22   9.5 GB: 5-of-22   9.2 GB: 6-of-22   9.3 GB: 7-of-22   9.9 GB: 8-of-22   9.5 GB: 9-of-22   9.2 GB: 10-of-22   9.3 GB: 11-of-22   9.9 GB: 12-of-22   9.5 GB: 13-of-22   9.2 GB: 14-of-22   9.3 GB: 15-of-22   9.9 GB: 16-of-22   9.5 GB: 17-of-22   9.2 GB: 18-of-22   9.3 GB: 19-of-22   9.9 GB: 20-of-22   9.5 GB: 21-of-22   5.0 GB: 22-of-22
Supported Languageszh en
Model ArchitectureAutoModel
Licenseapache-2.0
Transformers Version4.30.2
Tokenizer ClassFLMTokenizer
Beginning of Sentence Token<|endoftext|>
End of Sentence Token<|endoftext|>
Unk Token<|endoftext|>
Vocabulary Size100352
Activation Functiongelu_fast
Errorsreplace