divakantpandey/code-embed-v2

Code-Embed v2 (Fast Lane Update) 🚀

State-of-the-Art Code Embedding & Semantic Search Model

Developed by Divakant Pandey, code-embed-v2 is a premier code embedding model engineered for dense code retrieval, developer Q&A search, competitive algorithm matching, cross-language translation, and multi-turn instruction following.

---

🌟 Benchmark Performance: Yesterday vs. Fast Lane

Evaluating across all 17 official MTEB, CoIR, and CoREB benchmarks shows massive empirical gains across every single programming modality:

Benchmark Task	Category	Baseline (Yesterday)	Fast Lane (Updated)	Absolute Gain	Relative Gain
CodeSearchNetRetrieval	Text-to-Code Search	38.67%	89.16%	+50.49%	+130.6% (2.3×)
CodeSearchNetCCRetrieval	Code-to-Code Clone Search	15.58%	63.95%	+48.37%	+310.5% (4.1×)
CorebC2TRetrieval	Code-to-Text Documentation	22.32%	60.73%	+38.41%	+172.1%
CodeTransOceanContest	Algorithmic Translation	36.40%	68.14%	+31.74%	+87.2%
CodeFeedbackMT	Multi-Turn Context Matching	31.88%	61.64%	+29.76%	+93.4%
CorebC2CRetrieval	Code-to-Code Cross-Repo	2.87%	27.50%	+24.63%	+858.2% (9.6×)
CodeFeedbackST	Single-Turn Instruction Match	46.93%	66.80%	+19.87%	+42.3%
CorebT2CRetrieval	Text-to-Code Search	7.45%	26.71%	+19.26%	+258.5% (3.6×)
CorebC2TReranking	Code-to-Text Reranking	66.74%	84.55%	+17.81%	+26.7%
AppsRetrieval	Complex Problem Solving	3.06%	20.17%	+17.11%	+559.0% (6.6×)
SyntheticText2SQL	Natural Language to SQL	32.77%	49.38%	+16.61%	+50.7%
StackOverflowQA	Developer Q&A Matching	55.30%	71.31%	+16.01%	+29.0%
CorebC2CReranking	Code-to-Code Reranking	36.87%	51.54%	+14.67%	+39.8%
CosQA	Web Query Code Search	15.99%	29.97%	+13.98%	+87.4%
CorebT2CReranking	Text-to-Code Reranking	26.44%	36.81%	+10.37%	+39.2%
CodeTransOceanDL	Deep Learning Translation	34.39%	35.33%	+0.94%	+2.7%
COIRCodeSearchNetRetrieval	Multilingual Search	98.45%	94.22%	-4.23%	-4.3%
OVERALL MACRO-AVERAGE	17 Official Benchmarks	33.65%	55.17%	+21.52%	+63.9%

---

💻 Quickstart Usage

1. Basic Semantic Code Search with sentence-transformers

from sentence_transformers import SentenceTransformer

# Load updated model directly from Hugging Face Hub
model = SentenceTransformer("divakantpandey/code-embed-v2")

queries = [
    "binary search implementation",
    "fetch json from api endpoint",
    "goroutine channel worker pool"
]

code_snippets = [
    "def binary_search(arr, target):\n    left, right = 0, len(arr) - 1\n    while left <= right:\n        mid = (left + right) // 2\n        if arr[mid] == target: return mid\n        elif arr[mid] < target: left = mid + 1\n        else: right = mid - 1\n    return -1",
    "async function fetchJSON(url) {\n    const response = await fetch(url);\n    return await response.json();\n}",
    "func worker(id int, jobs <-chan int, results chan<- int) {\n    for j := range jobs {\n        results <- j * 2\n    }\n}"
]

# Generate normalized embeddings (1536-dim native)
query_embeddings = model.encode(queries, normalize_embeddings=True)
code_embeddings = model.encode(code_snippets, normalize_embeddings=True)

# Compute cosine similarity
similarity_matrix = query_embeddings @ code_embeddings.T
print(similarity_matrix)

2. Matryoshka Representation Learning (Adaptive Dimensions)

code-embed-v2 supports adaptive vector truncation. You can truncate embeddings down to 768, 512, 256, or 128 dimensions for reduced storage with minimal retrieval accuracy trade-off:

# Truncate to 768 dimensions (50% storage & compute reduction)
compact_query_embs = model.encode(queries, truncate_dim=768, normalize_embeddings=True)
compact_code_embs = model.encode(code_snippets, truncate_dim=768, normalize_embeddings=True)

similarity = compact_query_embs @ compact_code_embs.T

---

📜 License & Commercial Terms

This model and its associated weights are licensed under the Creative Commons Attribution-NonCommercial 4.0 International Public License (CC-BY-NC 4.0).

• Non-Commercial Use: Free for academic research, educational use, personal exploration, open-source evaluations, and non-commercial benchmarking with attribution.
• Attribution: Please attribute Divakant Pandey and reference the repository at https://github.com/divakantpandey/code-embed.
• Commercial Licensing: To license CodeEmbed v2 for commercial products, enterprise SaaS applications, closed-source systems, or proprietary developer tools, please contact:
  • Licensor / Maintainer: Divakant Pandey
  • Email: divakant01@gmail.com
  • GitHub: @divakantpandey