Sentence Similarity
sentence-transformers
ONNX
Safetensors
English
code
modernbert
feature-extraction
code-search
code-retrieval
bug-localization
Eval Results (legacy)
text-embeddings-inference
Instructions to use vedevpatel/CodeIntel-150M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- sentence-transformers
How to use vedevpatel/CodeIntel-150M with sentence-transformers:
from sentence_transformers import SentenceTransformer model = SentenceTransformer("vedevpatel/CodeIntel-150M") sentences = [ "That is a happy person", "That is a happy dog", "That is a very happy person", "Today is a sunny day" ] embeddings = model.encode(sentences) similarities = model.similarity(embeddings, embeddings) print(similarities.shape) # [4, 4] - Notebooks
- Google Colab
- Kaggle
Link the GitHub repo and add MCP setup for coding agents
Browse files
README.md
CHANGED
|
@@ -50,6 +50,9 @@ and CodeSearchNet Python. It is a fine-tune of
|
|
| 50 |
[granite-embedding-english-r2](https://huggingface.co/ibm-granite/granite-embedding-english-r2) (ModernBERT,
|
| 51 |
Apache-2.0). It runs locally, including as a 325 MB INT8 ONNX model, and needs no remote code.
|
| 52 |
|
|
|
|
|
|
|
|
|
|
| 53 |
| | |
|
| 54 |
|---|---|
|
| 55 |
| Parameters | 149M |
|
|
@@ -149,6 +152,20 @@ docstrings-TypeScript -0.002 [-0.010, +0.006]. It is about 1.7x faster on CPU.
|
|
| 149 |
**Hybrid retrieval.** Fusing dense scores with BM25 by reciprocal-rank fusion (dense weight 2, k = 60) adds about
|
| 150 |
+0.03 on issue localization.
|
| 151 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 152 |
## Training
|
| 153 |
|
| 154 |
- **Objective:** multi-positive InfoNCE (symmetric, temperature 0.05) with BM25-mined hard negatives and GradCache
|
|
|
|
| 50 |
[granite-embedding-english-r2](https://huggingface.co/ibm-granite/granite-embedding-english-r2) (ModernBERT,
|
| 51 |
Apache-2.0). It runs locally, including as a 325 MB INT8 ONNX model, and needs no remote code.
|
| 52 |
|
| 53 |
+
Training and evaluation code, the benchmarks and an MCP server for coding agents are at
|
| 54 |
+
[github.com/vedevpatel/codeintel-150m](https://github.com/vedevpatel/codeintel-150m).
|
| 55 |
+
|
| 56 |
| | |
|
| 57 |
|---|---|
|
| 58 |
| Parameters | 149M |
|
|
|
|
| 152 |
**Hybrid retrieval.** Fusing dense scores with BM25 by reciprocal-rank fusion (dense weight 2, k = 60) adds about
|
| 153 |
+0.03 on issue localization.
|
| 154 |
|
| 155 |
+
## Use with coding agents (MCP)
|
| 156 |
+
|
| 157 |
+
`codeintel mcp` gives Claude Code, Codex, Cursor and other MCP clients semantic search over the repository they are
|
| 158 |
+
working in: the agent passes an issue or a plain-language description and gets back ranked functions and classes with
|
| 159 |
+
file paths and line ranges. It runs locally over stdio, indexes the repository in the background and re-embeds only
|
| 160 |
+
changed code after edits.
|
| 161 |
+
|
| 162 |
+
```bash
|
| 163 |
+
# Claude Code (run in the repository)
|
| 164 |
+
claude mcp add codeintel -- uvx --from "codeintel[mcp] @ git+https://github.com/vedevpatel/codeintel-150m" codeintel mcp .
|
| 165 |
+
```
|
| 166 |
+
|
| 167 |
+
Setup for Codex and Cursor is in the [GitHub README](https://github.com/vedevpatel/codeintel-150m#use-with-coding-agents-mcp).
|
| 168 |
+
|
| 169 |
## Training
|
| 170 |
|
| 171 |
- **Objective:** multi-positive InfoNCE (symmetric, temperature 0.05) with BM25-mined hard negatives and GradCache
|