LLM EXPLORER 59,420 MODELS INDEXED

Internlm2 Math 20B by internlm

By internlm · 78 downloads
HF Repo ↗ Copy URL

Internlm2 Math 20B is an open-source language model by internlm. Features: 20b LLM, VRAM: 39.7GB, Context: 8K, License: other, LLM Explorer Score: 0.11.

  Conversational   Custom code   En   Internlm2   Math   Region:us   Safetensors   Sharded   Tensorflow   Zh
Model Card on HF 🤗: https://huggingface.co/internlm/internlm2-math-20b 

Internlm2 Math 20B Parameters and Internals

Model Type 
math reasoning, text generation
Use Cases 
Areas:
Bilingual Math, Math reasoning, Verification
Applications:
Math problem solving, Math theorem proving, Mathematical chain-of-thought verification, Lean code use
Primary Use Cases:
Math problem solving and theorem proving, Generating Lean code for simple reasoning tasks, Outcome, process, and Lean reward modeling
Limitations:
Jumping calculation steps, Performance variance in some contexts, Tendency to generate code for Chinese problems, Some Lean capabilities limited to GSM8K-like problems
Additional Notes 
The model integrates Lean language support, leveraging it for GSM8K and similar reasoning tests.
Supported Languages 
en (high), zh (high)
Training Details 
Data Sources:
InternLM2-Base, ~100B Math-related tokens, ~2M bilingual supervised data
Data Volume:
100 billion tokens
Methodology:
Continued pretraining and fine-tuning with specialized techniques such as minhash and exact number match to decontaminate possible test set leakage
Input Output 
Input Format:
Text input expected
Accepted Modalities:
text
Output Format:
Text
Performance Tips:
Use prompt instructions correctly for optimal results.
Release Notes 
Version:
InternLM2-Math-Base-7B
Date:
2024-01-23
Notes:
Pretrained checkpoint release.
Version:
InternLM2-Math-7B
Date:
2024-01-23
Notes:
SFT checkpoint release.
LLM NameInternlm2 Math 20B
Repository 🤗https://huggingface.co/internlm/internlm2-math-20b 
Model Size20b
Required VRAM39.7 GB
Updated2026-07-27
Maintainerinternlm
Model Typeinternlm2
Model Files  1.9 GB: 1-of-21   1.9 GB: 2-of-21   2.0 GB: 3-of-21   1.9 GB: 4-of-21   2.0 GB: 5-of-21   1.9 GB: 6-of-21   2.0 GB: 7-of-21   1.9 GB: 8-of-21   2.0 GB: 9-of-21   1.9 GB: 10-of-21   2.0 GB: 11-of-21   1.9 GB: 12-of-21   2.0 GB: 13-of-21   1.9 GB: 14-of-21   2.0 GB: 15-of-21   1.9 GB: 16-of-21   2.0 GB: 17-of-21   1.9 GB: 18-of-21   2.0 GB: 19-of-21   1.6 GB: 20-of-21   1.1 GB: 21-of-21
Supported Languagesen zh
Model ArchitectureInternLM2ForCausalLM
Licenseother
Context Length8192
Model Max Length8192
Transformers Version4.35.2
Is Biased0
Tokenizer ClassInternLM2Tokenizer
Padding Token</s>
Vocabulary Size92544
Torch Data Typebfloat16

Quantized Models of the Internlm2 Math 20B

Model
Likes
Downloads
VRAM
Internlm2 Math 20B AWQ0812 GB

Best Alternatives to Internlm2 Math 20B

Best Alternatives
Context / RAM
Downloads
Likes
Internlm2 5 20B256K / 39.7 GB15817
Internlm2 5 20B Chat32K / 39.7 GB764296
Internlm2 Chat 20B ExPO32K / 39.8 GB158211
Internlm2 20B32K / 39.7 GB2050359
Internlm2 Chat 20B32K / 39.7 GB2063588
Internlm2 Base 20B32K / 39.7 GB158048
...ternlm2 5 20B Chat Abliterated32K / 39.7 GB164
ChemLLM 20B Chat DPO32K / 40.6 GB3310
ChemLLM 20B Chat SFT32K / 40.6 GB445
Internlm Chatbode 20B32K / 39.8 GB41
Note: green Score (e.g. "73.2") means that the model is better than internlm/internlm2-math-20b.