README.md

metadata

license: apache-2.0
datasets:
  - Shuu12121/python-treesitter-filtered-datasetsV2
  - Shuu12121/javascript-treesitter-filtered-datasetsV2
  - Shuu12121/ruby-treesitter-filtered-datasetsV2
  - Shuu12121/go-treesitter-dedupe_doc-filtered-dataset
  - Shuu12121/java-treesitter-dedupe_doc-filtered-dataset
  - Shuu12121/rust-treesitter-filtered-datasetsV2
  - Shuu12121/php-treesitter-filtered-datasetsV2
  - Shuu12121/typescript-treesitter-filtered-datasetsV2
pipeline_tag: fill-mask
tags:
  - code
  - python
  - java
  - javascript
  - typescript
  - go
  - ruby
  - rust
  - php
language:
  - en
base_model:
  - Shuu12121/CodeModernBERT-Crow-v1-Pre

CodeModernBERT-Crow-v1.1🐦‍⬛

Model Details

• Model type: Bi-encoder architecture based on ModernBERT
• Architecture:
  • Hidden size: 768
  • Layers: 12
  • Attention heads: 12
  • Intermediate size: 3,072
  • Max position embeddings: 8,192
  • Local attention window size: 128
  • RoPE positional encoding: θ = 160,000
  • Local RoPE positional encoding: θ = 10,000
• Sequence length: up to 2,048 tokens for code and docstring inputs during pretraining

Pretraining

• Tokenizer: Custom BPE tokenizer trained for code and docstring pairs.

• Data: Functions and natural language descriptions extracted from GitHub repositories.

• Masking strategy: Two-phase pretraining.

  • Phase 1: Random Masked Language Modeling (MLM)
    30% of tokens in code functions are randomly masked and predicted using standard MLM.
  • Phase 2: Line-level Span Masking
    Inspired by SpanBERT, continued pretraining on the same data with span masking at line granularity:
    1. Convert input tokens back to strings.
    2. Detect newline tokens with regex and segment inputs by line.
    3. Exclude whitespace-only tokens from masking.
    4. Apply padding to align sequence lengths.
    5. Randomly mask 30% of tokens in each line segment and predict them.

• Pretraining hyperparameters:

  • Batch size: 16
  • Gradient accumulation steps: 16
  • Effective batch size: 256
  • Optimizer: AdamW
  • Learning rate: 5e-5
  • Scheduler: Cosine
  • Epochs: 3
  • Precision: Mixed precision (fp16) using transformers