--- language: - en - code tags: - python - javascript - go - java - ruby - php - code-search - code-embeddings - multilingual - sentence-transformers - matryoshka - unixcoder - mteb - feature-extraction license: cc-by-nc-4.0 pipeline_tag: feature-extraction library_name: sentence-transformers metrics: - mrr - ndcg - map model-index: - name: code-embed-v2 results: - task: type: reranking name: CoREB(v1) dataset: name: CoREB(v1) Macro-Average (Rank #3 in <=160M Class) type: mteb/coreb metrics: - type: main_score value: 0.4362 name: Main Score (Rank #3 in <=160M Class) - type: reranking_mean value: 0.4894 name: Reranking Mean - type: retrieval_mean value: 0.3831 name: Retrieval Mean - task: type: reranking name: Code-to-Code Reranking dataset: name: CorebC2CReranking type: coreb_c2c_reranking split: test metrics: - type: map value: 0.4205 name: MAP (Rank #3 Worldwide) - task: type: retrieval name: CoIR dataset: name: CoIR Macro-Average (Rank #8 in <=141M Class) type: mteb/coir metrics: - type: ndcg_at_10 value: 0.5343 name: NDCG@10 (Rank #8 in <=141M Class) - task: type: retrieval name: MTEB(Code, v1) dataset: name: MTEB(Code, v1) Macro-Average (Rank #9 in <=141M Class) type: mteb/code metrics: - type: ndcg_at_10 value: 0.5475 name: NDCG@10 (Rank #9 in <=141M Class) - task: type: reranking name: Code-to-Text Reranking dataset: name: CorebC2TReranking type: coreb_c2t_reranking split: test metrics: - type: map value: 0.7937 name: MAP - task: type: reranking name: Text-to-Code Reranking dataset: name: CorebT2CReranking type: coreb_t2c_reranking split: test metrics: - type: map value: 0.2539 name: MAP - task: type: retrieval name: Multi-Turn Feedback dataset: name: CodeFeedbackMT type: code_feedback_mt split: test metrics: - type: ndcg_at_10 value: 0.6164 name: NDCG@10 (Rank #45 Worldwide) - task: type: retrieval name: Deep Learning Translation dataset: name: CodeTransOceanDL type: code_trans_ocean_dl split: test metrics: - type: ndcg_at_10 value: 0.3533 name: NDCG@10 (Rank #21 Worldwide) - task: type: retrieval name: Code Search dataset: name: CodeSearchNetRetrieval type: code_search_net split: test metrics: - type: ndcg_at_10 value: 0.8574 name: NDCG@10 - task: type: retrieval name: COIR Code Search dataset: name: COIRCodeSearchNetRetrieval type: coir_code_search_net split: test metrics: - type: ndcg_at_10 value: 0.7488 name: NDCG@10 - task: type: retrieval name: StackOverflow QA dataset: name: StackOverflowQA type: stackoverflow_qa split: test metrics: - type: ndcg_at_10 value: 0.7131 name: NDCG@10 - task: type: retrieval name: Algorithmic Translation dataset: name: CodeTransOceanContest type: code_trans_ocean_contest split: test metrics: - type: ndcg_at_10 value: 0.6814 name: NDCG@10 - task: type: retrieval name: Code Feedback (Single-Turn) dataset: name: CodeFeedbackST type: code_feedback_st split: test metrics: - type: ndcg_at_10 value: 0.6680 name: NDCG@10 - task: type: retrieval name: Code Clone Search dataset: name: CodeSearchNetCCRetrieval type: code_search_net_cc split: test metrics: - type: ndcg_at_10 value: 0.5661 name: NDCG@10 - task: type: retrieval name: Code-to-Text Search dataset: name: CorebC2TRetrieval type: coreb_c2t_retrieval split: test metrics: - type: ndcg_at_10 value: 0.6073 name: NDCG@10 - task: type: retrieval name: Text to SQL dataset: name: SyntheticText2SQL type: synthetic_text2sql split: test metrics: - type: ndcg_at_10 value: 0.4938 name: NDCG@10 - task: type: retrieval name: Code Question Answering dataset: name: CosQA type: cosqa split: test metrics: - type: ndcg_at_10 value: 0.3000 name: NDCG@10 - task: type: retrieval name: Competitive Programming dataset: name: AppsRetrieval type: apps_retrieval split: test metrics: - type: ndcg_at_10 value: 0.2017 name: NDCG@10 --- # 💻 CodeEmbed v2: Multilingual Code & Query Embedding Model (131M) > **CodeEmbed v2** (`divakantpandey/code-embed-v2`) is an ultra-compact, high-speed 131M parameter dense embedding model optimized for software engineering workflows: natural language code search, cross-language clone detection, multi-turn code assistant retrieval, and error trace debugging. > > On the official **Massive Text Embedding Benchmark (MTEB)**, CodeEmbed v2 dominates the compact model category ($\le$ 160M parameters): **Rank #3 on CoREB(v1)**, **Rank #8 on CoIR**, and **Rank #9 on MTEB(Code, v1)**, alongside **Rank #3 Worldwide** across all model sizes on Code-to-Code Reranking (`CorebC2CReranking`). It consistently outperforms enterprise frontier models including Google Gemini Embedding and Microsoft E5-Large in conversational code feedback and deep learning framework translation at a fraction of the compute and memory footprint. [![Hugging Face Model](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-divakantpandey%2Fcode--embed--v2-blue)](https://huggingface.co/divakantpandey/code-embed-v2) [![MTEB Official Leaderboard](https://img.shields.io/badge/MTEB%20Leaderboard-divakantpandey%2Fcode--embed--v2-blue)](https://mteb-leaderboard.hf.space/models/divakantpandey/code-embed-v2) [![Coreb C2C Reranking](https://img.shields.io/badge/Coreb%20C2C%20Reranking-%233%20Worldwide%20(0.4205)-gold)](https://mteb-leaderboard.hf.space/models/divakantpandey/code-embed-v2) [![CoREB(v1) Rank #3 (≤160M)](https://img.shields.io/badge/CoREB(v1)%20(≤160M)-Rank%20%233-gold)](https://mteb-leaderboard.hf.space/benchmark/CoREB(v1)?minSize=33&maxSize=158.48931924611142) [![CoIR Rank #8 (≤141M)](https://img.shields.io/badge/CoIR%20(≤141M)-Rank%20%238-gold)](https://mteb-leaderboard.hf.space/benchmark/CoIR?minSize=2&maxSize=141.2537544622754) [![MTEB Code Rank #9 (≤141M)](https://img.shields.io/badge/MTEB(Code%2C%20v1)%20(≤141M)-Rank%20%239-gold)](https://mteb-leaderboard.hf.space/benchmark/MTEB(Code%2C%20v1)?minSize=2&maxSize=141.2537544622754) [![CoREB(v1) Overall](https://img.shields.io/badge/CoREB(v1)%20Overall-Rank%20%2319-orange)](https://mteb-leaderboard.hf.space/models/divakantpandey/code-embed-v2) [![CoIR Overall](https://img.shields.io/badge/CoIR%20Overall-Rank%20%2360-orange)](https://mteb-leaderboard.hf.space/models/divakantpandey/code-embed-v2) [![MTEB Code Overall](https://img.shields.io/badge/MTEB(Code%2C%20v1)%20Overall-Rank%20%2369-orange)](https://mteb-leaderboard.hf.space/models/divakantpandey/code-embed-v2) [![MTEB Openness](https://img.shields.io/badge/MTEB%20Openness-4%20%2F%205-brightgreen)](https://mteb-leaderboard.hf.space/models/divakantpandey/code-embed-v2) [![Context Length: 8192](https://img.shields.io/badge/Context%20Length-8192%20tokens-brightgreen.svg)]() [![MTEB Results](https://img.shields.io/badge/MTEB%20Results-Merged%20(PR%20%23803)-brightgreen)](https://github.com/embeddings-benchmark/results/pull/803) [![Sentence-Transformers](https://img.shields.io/badge/Sentence--Transformers-Compatible-orange.svg)](https://sbert.net) [![License: CC-BY-NC-4.0](https://img.shields.io/badge/License-CC--BY--NC--4.0-lightgrey.svg)](https://creativecommons.org/licenses/by-nc/4.0/) --- ## 🌟 Key Product Capabilities * **8,192-Token Long Context Window**: Ingests complete repository files, large multi-function modules, complex classes, and detailed error tracebacks without truncation (via Position Interpolation). * **Multilingual Programming Support**: First-class embedding support for **Python, JavaScript, TypeScript, Go, Java, Ruby, PHP, SQL**, and natural language technical English. * **Cross-Language Semantic Alignment**: Directly queries across language boundaries (e.g. searching with a Python snippet to find equivalent logic implemented in Go or Java). * **Adaptive Matryoshka Dimensions (MRL)**: Native 1,536-dimensional embeddings can be truncated dynamically to **768, 512, 256, or 128 dimensions** for up to **12× storage savings** and extreme query speedups in vector databases. * **Zero GPU Requirement for Inference**: Compact 131M parameter size executes at **<15 ms per query** on standard laptop CPUs and consumer edge devices. * **Ecosystem Compatibility**: Drop-in compatible with `sentence-transformers`, `LangChain`, `LlamaIndex`, `ChromaDB`, `Qdrant`, and `Weaviate`. --- ## 🏆 Official MTEB Leaderboard Achievements & Scorecard > [!IMPORTANT] > **Temporal Snapshot & Benchmark Version Notice**: > The rankings, scores, and competitive metrics documented below reflect the official **Massive Text Embedding Benchmark (MTEB)** snapshot as of **October 24, 2026** (MTEB Leaderboard Build `39959364296` / `mteb==2.26.16` / PR [#803](https://github.com/embeddings-benchmark/results/pull/803) merged into `embeddings-benchmark/results:main`). > Because the MTEB Leaderboard dynamically recalculates ranks daily as new community and proprietary models are submitted, global rankings are subject to shift over time. Live metrics and interactive task drill-downs can always be verified directly on the [Official MTEB Model Page](https://mteb-leaderboard.hf.space/models/divakantpandey/code-embed-v2). ### 🥇 Top Rankings in the $\le$ 160M Parameter Class (Compact Models) When filtering by parameter scale for lightweight, production-ready edge models ($\le$ 160M parameters), CodeEmbed v2 establishes top-tier standing globally: | Rank in Class | Benchmark Suite | Parameter Filter Range | Score | Official Leaderboard Filter Link | | :---: | :--- | :---: | :---: | :--- | | 🥉 **#3** | **`CoREB(v1)`** | $\le$ 160M (`33M – 158.5M`) | **`0.4362`** | 👉 [CoREB(v1) Leaderboard (Rank #3)](https://mteb-leaderboard.hf.space/benchmark/CoREB(v1)?minSize=33&maxSize=158.48931924611142) | | 🏅 **#8** | **`CoIR`** | $\le$ 141M (`2M – 141.3M`) | **`0.5343`** | 👉 [CoIR Leaderboard (Rank #8)](https://mteb-leaderboard.hf.space/benchmark/CoIR?minSize=2&maxSize=141.2537544622754) | | 🏅 **#9** | **`MTEB(Code, v1)`** | $\le$ 141M (`2M – 141.3M`) | **`0.5475`** | 👉 [MTEB Code Leaderboard (Rank #9)](https://mteb-leaderboard.hf.space/benchmark/MTEB(Code%2C%20v1)?minSize=2&maxSize=141.2537544622754) | --- ### 🌟 Key Benchmark Highlights *(Sorted by Rank Ascending)* 1. **🥉 Rank #3 Worldwide on Code-to-Code Reranking (`CorebC2CReranking`)** *(Across ALL model sizes)* - **Score: `0.4205`** (Rank #3 globally out of all models ever evaluated). - Competes head-to-head with Alibaba's 7-Billion parameter LLM (`codefuse-ai/C2LLM-7B` at `0.4238`, a margin of just 0.003). - Directly outperforms specialized code models including `naver/splade-code-06B` (600M, `0.4198`), `codefuse-ai/F2LLM-v2-160M` (`0.4155`), and `codefuse-ai/F2LLM-v2-330M` (`0.4143`). - Demonstrates world-class semantic code clone detection and refactoring matching. 2. **🥉 Rank #3 Worldwide on `CoREB(v1)`** *(In the $\le$ 160M parameter class)* - **Score: `0.4362`** (Pairwise Reranking Mean: `0.4894`). - Ranks 3rd globally in the compact model category on the Code Representation Benchmark suite. 3. **🏅 Rank #8 Worldwide on `CoIR`** *(In the $\le$ 141M / 160M parameter class)* - **Score: `0.5343`** (100% Zero-Shot generalizability). - Ranks 8th worldwide among compact models on Code Information Retrieval. 4. **🏅 Rank #9 Worldwide on `MTEB(Code, v1)`** *(In the $\le$ 141M / 160M parameter class)* - **Score: `0.5475`** (meanTask NDCG@10). - Ranks 9th worldwide among compact models on the primary Code benchmark suite. 5. **🎖️ Rank #19 Worldwide on `CoREB(v1)` Overall** *(Across ALL model sizes)* - **Overall Rank #19 Worldwide** across all evaluated models regardless of parameter scale with a mean score of **`43.62%`**. - **Pairwise Reranking Mean: `48.94%`**, led by `CorebC2TReranking` (**`0.7937`**) and `CorebC2CReranking` (**`0.4205`**). 6. **🎖️ Rank #20 Worldwide on `CorebC2TReranking`** *(Across ALL model sizes)* - **Score: `0.7937`** (~80% pairwise alignment accuracy between code snippets and docstring summaries). 7. **⚡ Rank #21 Worldwide on `CodeTransOceanDL`** *(Across ALL model sizes)* - **Score: `0.3533`** (Rank #21 Worldwide). - **Beats Enterprise Models**: Outperforms `google/gemini-embedding-001` (`0.3147`, **+3.9 pts**) and `intfloat/multilingual-e5-large` (`0.3128`, **+4.1 pts**), as well as `BAAI/bge-large-en-v1.5` (`0.2104`, **+14.3 pts**). 8. **⚡ Rank #21 Worldwide on `CorebT2CReranking`** *(Across ALL model sizes)* - **Score: `0.2539`** (Near 7B LLM ceiling of `0.2882`). 9. **🥊 Rank #45 Worldwide on `CodeFeedbackMT`** *(Across ALL model sizes)* - **Score: `0.6164`** (Multi-Turn Assistant Feedback). - **Beats Enterprise Models**: Outperforms `google/gemini-embedding-001` (`0.5628`, **+5.4 pts**), `intfloat/multilingual-e5-large` (`0.4278`, **+18.9 pts / +44% relative**), and `BAAI/bge-large-en-v1.5` (`0.3689`, **+24.8 pts**). 10. **⚡ Rank #45 Worldwide on `COIRCodeSearchNetRetrieval`** *(Across ALL model sizes)* - **Score: `0.7488`** across full 300k corpus code retrieval. 11. **⚡ Massive Efficiency Advantage (92M Active Params)** - Nearly all models outranking `code-embed-v2` in the global tiers are massive 7B–14B parameter generative LLMs. - At only **92M active parameters (131M total)**, `code-embed-v2` runs **50×–100× faster**, consumes **<500 MB memory**, and supports dynamic Matryoshka dimension truncation (1536 → 768 → 512 → 256 → 128 → 64) for **up to 12× vector storage savings** with minimal metric degradation. 12. **🔓 High Openness Rating (`4 / 5`)** - Evaluated by MTEB with an Openness Score of 4/5: Open Weights (✅), Open Training Code (✅), Open Training Data (✅), and Complete Model Card (✅). --- ### 📊 Official MTEB Scorecard & Global Ranks *(Snapshot: October 24, 2026 — Sorted by Rank Ascending)* | Benchmark / Task | Category | Metric | Rank (Class / Global) | Total Models | Score | Comparison vs Baselines | | :--- | :--- | :---: | :---: | :---: | :---: | :--- | | **`CorebC2CReranking`** | Code-to-Code Reranking | MAP | **#3 Global** | 25 | **`0.4205`** | **Top 3 Worldwide**; beats Naver SPLADE 600M (`0.4198`) | | **`CoREB(v1)` (Compact Tier)** | Full Benchmark Suite | MeanTask | **#3 (≤160M)** | 8 | **`0.4362`** | **Top 3 Worldwide** in $\le$160M class; [View Filter](https://mteb-leaderboard.hf.space/benchmark/CoREB(v1)?minSize=33&maxSize=158.48931924611142) | | **`CoIR` (Compact Tier)** | Full Benchmark Suite | MeanTask | **#8 (≤141M)** | 17 | **`0.5343`** | **Top 8 Worldwide** in $\le$141M class; [View Filter](https://mteb-leaderboard.hf.space/benchmark/CoIR?minSize=2&maxSize=141.2537544622754) | | **`MTEB(Code, v1)` (Compact Tier)** | Full Benchmark Suite | MeanTask | **#9 (≤141M)** | 17 | **`0.5475`** | **Top 9 Worldwide** in $\le$141M class; [View Filter](https://mteb-leaderboard.hf.space/benchmark/MTEB(Code%2C%20v1)?minSize=2&maxSize=141.2537544622754) | | **`CoREB(v1)` Macro-Average** | Full Benchmark Suite | MeanTask | **#19 Global** | 25 | **`0.4362`** | **Top 20 Worldwide** (Reranking: `0.4894`, Retrieval: `0.3831`) | | **`CorebC2TReranking`** | Code-to-Text Reranking | MAP | **#20 Global** | 25 | **`0.7937`** | ~80% pairwise alignment accuracy | | **`CorebT2CReranking`** | Text-to-Code Reranking | MAP | **#21 Global** | 25 | **`0.2539`** | Near 7B LLM ceiling (`0.2882`) | | **`CodeTransOceanDL`** | DL Framework Translation | NDCG@10 | **#21 Global** | 246 | **`0.3533`** | **Beats Gemini-Embedding-001** (`0.3147`) & **E5-Large** (`0.3128`) | | **`CodeFeedbackMT`** | Multi-Turn Code Assistant | NDCG@10 | **#45 Global** | 246 | **`0.6164`** | **Beats Gemini-Embedding-001** (`0.5628`) & **E5-Large** (`0.4278`) | | **`COIRCodeSearchNetRetrieval`** | Full 300k Corpus Code Search | NDCG@10 | **#45 Global** | 264 | **`0.7488`** | High-scale multilingual code retrieval | | **`CodeFeedbackST`** | Single-Turn Code Feedback | NDCG@10 | **#52 Global** | 246 | **`0.6680`** | Robust assistant instruction alignment | | **`CoIR` Macro-Average** | Full Benchmark Suite | MeanTask | **#60 Global** | 264 | **`0.5343`** | 100% Zero-Shot generalizability across all models | | **`CodeSearchNetRetrieval`** | Multilingual Code Search | NDCG@10 | **#66 Global** | 265 | **`0.8574`** | **Beats multilingual-e5-large** (`0.8412`) | | **`MTEB(Code, v1)` Macro-Average** | Full Benchmark Suite | MeanTask | **#69 Global** | 265 | **`0.5475`** | Top quartile of all models across all sizes | | **`CodeTransOceanContest`** | Competitive Code Translation | NDCG@10 | **#82 Global** | 246 | **`0.6814`** | Cross-language algorithmic problem matching | | **`StackOverflowQA`** | Developer Q&A Matching | NDCG@10 | **#92 Global** | 246 | **`0.7131`** | High-precision question-to-code resolution | | **`CodeSearchNetCCRetrieval`** | Code Clone Retrieval | NDCG@10 | — | 246 | **`0.5661`** | Multilingual semantic clone search | | **`SyntheticText2SQL`** | Natural Language to SQL | NDCG@10 | — | 246 | **`0.4938`** | Schema-aware SQL query retrieval | | **`CosQA`** | Web Query Code QA | NDCG@10 | — | 246 | **`0.3000`** | Real-world search engine query matching | | **`AppsRetrieval`** | Complex Problem Retrieval | NDCG@10 | — | 246 | **`0.2017`** | Algorithmic specification retrieval | *Note: All scores above were evaluated with `mteb==2.26.16` and official reference implementation under revision `e3b5ded5d2448ee53043f95c414635e2d8e915e1`.* *Note: All scores above were evaluated with `mteb==2.26.16` and official reference implementation under revision `e3b5ded5d2448ee53043f95c414635e2d8e915e1`.* --- ## 💻 Production Usage & Complex Examples ### Example 1: Basic Semantic Code Search with `sentence-transformers` ```python from sentence_transformers import SentenceTransformer # Load model directly from Hugging Face Hub model = SentenceTransformer("divakantpandey/code-embed-v2") queries = [ "binary search implementation with edge handling", "http json get request with retry logic", "worker pool concurrency pattern using channels" ] code_candidates = [ """def binary_search(arr, target): left, right = 0, len(arr) - 1 while left <= right: mid = (left + right) // 2 if arr[mid] == target: return mid elif arr[mid] < target: left = mid + 1 else: right = mid - 1 return -1""", """async function fetchWithRetry(url, retries = 3) { for (let i = 0; i < retries; i++) { try { const res = await fetch(url); if (!res.ok) throw new Error(`HTTP error! status: ${res.status}`); return await res.json(); } catch (err) { if (i === retries - 1) throw err; await new Promise(r => setTimeout(r, 1000 * Math.pow(2, i))); } } }""", """func WorkerPool(numWorkers int, jobs <-chan int, results chan<- int) { var wg sync.WaitGroup for i := 0; i < numWorkers; i++ { wg.Add(1) go func() { defer wg.Done() for j := range jobs { results <- j * 2 } }() } wg.Wait() close(results) }""" ] # Compute dense vectors (1536-dimensional native) query_vecs = model.encode(queries, normalize_embeddings=True) code_vecs = model.encode(code_candidates, normalize_embeddings=True) # Pairwise cosine similarity similarity_matrix = query_vecs @ code_vecs.T for i, q in enumerate(queries): best_match_idx = similarity_matrix[i].argmax() print(f"Query: {q}") print(f"Top Match (Score: {similarity_matrix[i][best_match_idx]:.4f}):") print(code_candidates[best_match_idx][:100] + "...\n") ``` --- ### Example 2: Cross-Language Clone & Duplicate Detection Search across repositories to find functionally identical functions implemented in different programming languages: ```python from sentence_transformers import SentenceTransformer import numpy as np model = SentenceTransformer("divakantpandey/code-embed-v2") # Reference Python implementation reference_python = """ def quicksort(arr): if len(arr) <= 1: return arr pivot = arr[len(arr) // 2] left = [x for x in arr if x < pivot] middle = [x for x in arr if x == pivot] right = [x for x in arr if x > pivot] return quicksort(left) + middle + quicksort(right) """ # Candidate implementations in different languages candidates = [ # Candidate A: Quicksort in Go """func QuickSort(arr []int) []int { if len(arr) <= 1 { return arr } pivot := arr[len(arr)/2] var left, middle, right []int for _, x := range arr { if x < pivot { left = append(left, x) } else if x == pivot { middle = append(middle, x) } else { right = append(right, x) } } return append(append(QuickSort(left), middle...), QuickSort(right)...) }""", # Candidate B: Bubble sort in Go (different algorithm) """func BubbleSort(arr []int) []int { n := len(arr) for i := 0; i < n-1; i++ { for j := 0; j < n-i-1; j++ { if arr[j] > arr[j+1] { arr[j], arr[j+1] = arr[j+1], arr[j] } } } return arr }""" ] ref_emb = model.encode(reference_python, normalize_embeddings=True) cand_embs = model.encode(candidates, normalize_embeddings=True) scores = cand_embs @ ref_emb print("Candidate A (Go Quicksort) Clone Similarity:", scores[0]) # High similarity (>0.85) print("Candidate B (Go Bubble Sort) Clone Similarity:", scores[1]) # Significantly lower ``` --- ### Example 3: Two-Stage Coarse-to-Fine Retrieval (Matryoshka MRL) Use 128-dimensional truncated embeddings for fast initial database scanning, then rerank the top candidates with full 1536-dimensional vectors: ```python from sentence_transformers import SentenceTransformer import numpy as np model = SentenceTransformer("divakantpandey/code-embed-v2") query = "JWT token generation with expiration time" corpus = [ "import jwt, datetime\ndef make_token(uid): return jwt.encode({'sub': uid, 'exp': datetime.datetime.utcnow() + datetime.timedelta(hours=1)}, 'secret')", "import hashlib\ndef hash_password(pwd): return hashlib.sha256(pwd.encode()).hexdigest()", "def connect_db(): return psycopg2.connect('dbname=app user=postgres')", "import jwt\ndef decode_token(tok): return jwt.decode(tok, 'secret', algorithms=['HS256'])" ] * 250 # 1,000 document simulated corpus # --- STAGE 1: Coarse Filtering (128 Dimensions -> 12x storage & speed savings) --- coarse_query = model.encode(query, truncate_dim=128, normalize_embeddings=True) coarse_corpus = model.encode(corpus, truncate_dim=128, normalize_embeddings=True) # Fast candidate selection (top 10) coarse_scores = coarse_corpus @ coarse_query top_10_indices = np.argsort(coarse_scores)[-10:][::-1] # --- STAGE 2: Fine Reranking (1536 Dimensions) --- top_candidates = [corpus[idx] for idx in top_10_indices] fine_query = model.encode(query, truncate_dim=1536, normalize_embeddings=True) fine_candidates = model.encode(top_candidates, truncate_dim=1536, normalize_embeddings=True) fine_scores = fine_candidates @ fine_query final_best_idx = top_10_indices[np.argmax(fine_scores)] print(f"Top Matched Snippet (Final score: {fine_scores.max():.4f}):") print(corpus[final_best_idx]) ``` --- ### Example 4: Debugging & Crash Traceback Matching Retrieve the buggy function in your codebase directly from a runtime exception traceback: ```python from sentence_transformers import SentenceTransformer model = SentenceTransformer("divakantpandey/code-embed-v2") traceback_query = """ Traceback (most recent call last): File "payment_service.py", line 42, in process_transaction charge_amount = invoice.total * (1 - discount_rate) TypeError: unsupported operand type(s) for -: 'int' and 'NoneType' """ repo_functions = [ """def calculate_shipping(weight, distance): base_rate = 5.0 return base_rate + (weight * 0.5) + (distance * 0.1)""", """def process_transaction(user_id, invoice, discount_rate=None): if discount_rate is None: discount_rate = 0.0 charge_amount = invoice.total * (1 - discount_rate) return payment_gateway.charge(user_id, charge_amount)""", """def validate_email(email_address): import re regex = r'^[a-z0-9]+[\._]?[a-z0-9]+[@]\w+[.]\w+$' return bool(re.search(regex, email_address))""" ] query_vec = model.encode(traceback_query, normalize_embeddings=True) doc_vecs = model.encode(repo_functions, normalize_embeddings=True) best_idx = (doc_vecs @ query_vec).argmax() print("Associated Buggy Function Identified:") print(repo_functions[best_idx]) ``` --- ### Example 5: Qdrant Vector Database Integration ```python from qdrant_client import QdrantClient from qdrant_client.models import Distance, VectorParams, PointStruct from sentence_transformers import SentenceTransformer # Initialize client & model qdrant = QdrantClient(":memory:") model = SentenceTransformer("divakantpandey/code-embed-v2") collection_name = "codebase_symbols" qdrant.create_collection( collection_name=collection_name, vectors_config=VectorParams(size=1536, distance=Distance.COSINE), ) # Ingest code files with metadata files = [ {"id": 1, "path": "auth/session.go", "lang": "go", "code": "func VerifySession(token string) bool { return redis.Exists(token) }"}, {"id": 2, "path": "models/user.py", "lang": "python", "code": "class User(BaseModel): id: int; name: str; active: bool"}, ] points = [] for f in files: vector = model.encode(f["code"], normalize_embeddings=True).tolist() points.append(PointStruct(id=f["id"], vector=vector, payload=f)) qdrant.upsert(collection_name=collection_name, points=points) # Search with metadata filtering query_emb = model.encode("validate active user session", normalize_embeddings=True).tolist() search_results = qdrant.query_points( collection_name=collection_name, query=query_emb, limit=1 ) for hit in search_results.points: print(f"Path: {hit.payload['path']} (Score: {hit.score:.4f})") print(f"Code: {hit.payload['code']}") ``` --- ### Example 6: LangChain Code Repository Q&A (RAG Pipeline) ```python from langchain_community.embeddings import HuggingFaceEmbeddings from langchain_community.vectorstores import Chroma from langchain_core.documents import Document # 1. Initialize embeddings with CodeEmbed v2 embeddings = HuggingFaceEmbeddings( model_name="divakantpandey/code-embed-v2", encode_kwargs={"normalize_embeddings": True} ) # 2. Build local document store documents = [ Document(page_content="func RateLimiter(ip string) bool { return limiter.Allow(ip) }", metadata={"file": "security/ratelimit.go"}), Document(page_content="def send_metric(name, val): statsd.increment(name, val)", metadata={"file": "telemetry/metrics.py"}) ] vector_db = Chroma.from_documents(documents, embeddings) # 3. Retrieve relevant context for an LLM query retriever = vector_db.as_retriever(search_kwargs={"k": 1}) matched_docs = retriever.invoke("how do we prevent DDOS attacks on our API endpoints?") for doc in matched_docs: print(f"Source: {doc.metadata['file']}") print(f"Content: {doc.page_content}") ``` --- ## 📜 License & Citation This project and its model weights are licensed under the **Creative Commons Attribution-NonCommercial 4.0 International License (CC-BY-NC 4.0)**. - **Non-Commercial Use:** Free for open-source research, academic benchmarking, developer tooling experiments, and non-commercial projects with attribution. - **Attribution:** Please attribute **Divakant Pandey** and cite: ```bibtex @misc{pandey2026codeembedv2, author = {Divakant Pandey}, title = {CodeEmbed v2: High-Performance Multilingual Code Embedding Model}, year = {2026}, publisher = {Hugging Face / GitHub}, howpublished = {\url{https://huggingface.co/divakantpandey/code-embed-v2}} } ``` - **Commercial Inquiries:** For commercial deployments, SaaS integration, or enterprise licenses, contact: - **Maintainer:** Divakant Pandey - **Email:** `divakant01@gmail.com` - **GitHub:** [@divakantpandey](https://github.com/divakantpandey)