Upload model card README for vtx-embed-1M

README.md

@@ -0,0 +1,194 @@
+---
+language:
+- en
+license: mit
+library_name: tokenizers
+tags:
+- sentence-similarity
+- feature-extraction
+- embeddings
+- rag
+- quantized
+- 4-bit
+- matryoshka
+- ultra-lightweight
+- code-search
+- retrieval
+pipeline_tag: feature-extraction
+metrics:
+- spearman_cosine
+model-index:
+- name: vtx-embed-1M
+  results:
+  - task:
+      type: sts
+      name: Semantic Textual Similarity
+    dataset:
+      name: STSBenchmark
+      type: mteb/stsbenchmark-sts
+    metrics:
+    - type: cosine_spearman
+      value: 0.7149
+  - task:
+      type: sts
+      name: Semantic Textual Similarity
+    dataset:
+      name: SICK-R
+      type: mteb/sickr-sts
+    metrics:
+    - type: cosine_spearman
+      value: 0.5841
+  - task:
+      type: classification
+      name: Classification
+    dataset:
+      name: Banking77Classification
+      type: mteb/banking77
+    metrics:
+    - type: accuracy
+      value: 0.6420
+  - task:
+      type: classification
+      name: Classification
+    dataset:
+      name: AmazonCounterfactualClassification
+      type: mteb/amazon_counterfactual_classification
+    metrics:
+    - type: accuracy
+      value: 0.6185
+  - task:
+      type: clustering
+      name: Clustering
+    dataset:
+      name: TwentyNewsgroupsClustering
+      type: mteb/twentynewsgroups-clustering
+    metrics:
+    - type: v_measure
+      value: 0.2512
+  - task:
+      type: clustering
+      name: Clustering
+    dataset:
+      name: RedditClustering
+      type: mteb/reddit-clustering
+    metrics:
+    - type: v_measure
+      value: 0.3140
+---
+
+<div align="center">
+
+# 🚀 vtx-embed-1M
+
+**Sub-megabyte pure data-free distilled embedding model.**
+Native 4-Bit quantization · 0.57 MB RAM · 1.05M Parameters · 1.14M Tokens/sec CPU Latency
+
+[![HuggingFace](https://img.shields.io/badge/🤗%20HuggingFace-VTXAI%2Fvtx--embed--1M-blue)](https://huggingface.co/VTXAI/vtx-embed-1M)
+[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT)
+[![Python 3.8+](https://img.shields.io/badge/Python-3.8%2B-blue)](https://python.org)
+
+</div>
+
+---
+
+## 📄 Model Details
+
+| Property | Value |
+| :--- | :--- |
+| **Model Type** | Native 4-Bit Data-Less Distilled Embedding (LF4) |
+| **Total Parameters** | **1.05M (1,048,576 parameters)** |
+| **Tensor Storage Format** | `lf4` — 4-bit per-block with FP16 scale + zero |
+| **In-RAM Memory** | **0.57 MB (570 KB)** |
+| **On-Disk Size** | **0.57 MB (570 KB)** |
+| **Vocabulary Size** | 16,384 |
+| **Max Sequence Length** | 512 tokens |
+| **Output Dimensions** | 64 *(Matryoshka: also 32)* |
+| **Pooling** | SIF IDF-weighted + PC-1 removal |
+| **Distillation Type** | 100% Data-Free SVD Subspace Reconstruction |
+| **License** | MIT |
+
+---
+
+## 📊 Comparative MTEB & STS Benchmark Results
+
+Comparing **vtx-embed-1M** side-by-side with **vtx-embed-7M**, **MiniLM-L6-v2**, and **bge-small-en-v1.5**:
+
+### Semantic Textual Similarity (STS)
+
+| Dataset | Metric | **vtx-embed-1M (0.57 MB)** | vtx-embed-7M (4.72 MB) | MiniLM-L6-v2 (90 MB) | bge-small-en-v1.5 (134 MB) |
+| :--- | :---: | :---: | :---: | :---: | :---: |
+| **STSBenchmark** | Spearman ρ | **0.7149** | 0.7918 | 0.8284 | 0.8278 |
+| **SICK-R** | Spearman ρ | **0.5841** | 0.6294 | 0.7572 | 0.7460 |
+
+### Classification
+
+| Dataset | Metric | **vtx-embed-1M (0.57 MB)** | vtx-embed-7M (4.72 MB) | MiniLM-L6-v2 (90 MB) | bge-small-en-v1.5 (134 MB) |
+| :--- | :---: | :---: | :---: | :---: | :---: |
+| **Banking77Classification** | Accuracy | **0.6420** | 0.7043 | 0.7451 | 0.7884 |
+| **AmazonCounterfactualClassification** | Accuracy | **0.6185** | 0.6679 | 0.7371 | 0.7279 |
+
+### Clustering
+
+| Dataset | Metric | **vtx-embed-1M (0.57 MB)** | vtx-embed-7M (4.72 MB) | MiniLM-L6-v2 (90 MB) | bge-small-en-v1.5 (134 MB) |
+| :--- | :---: | :---: | :---: | :---: | :---: |
+| **TwentyNewsgroupsClustering** | V-Measure | **0.2512** | 0.2936 | 0.3529 | 0.4419 |
+| **RedditClustering** | V-Measure | **0.3140** | 0.3880 | 0.4342 | 0.5376 |
+
+---
+
+## ⚡ Efficiency & Throughput Benchmark
+
+*Evaluated directly on CPU execution over 1,158 codebase files (900k+ tokens):*
+
+| Model | RAM Size | Throughput | Single File Latency | Tool Routing Accuracy |
+| :--- | :-: | :-: | :-: | :-: |
+| **vtx-embed-1M (64-dim)** | **0.57 MB** | **1,139,267 Tokens/sec** | **0.682 ms** | **100.0% (20/20)** |
+| **vtx-embed-7M (256-dim)** | **4.72 MB** | 878,578 Tokens/sec | 0.841 ms | **100.0% (20/20)** |
+| LiquidAI/LFM2.5-Embedding-350M | ~700 MB | ~137 texts/sec | 7.30 ms | — |
+| sentence-transformers/all-MiniLM-L6-v2 | 90 MB | ~80 texts/sec | 12.4 ms | — |
+
+> **vtx-embed-1M is 157× smaller than MiniLM-L6-v2 while maintaining a 100% Agent Tool Search Accuracy.**
+
+---
+
+## 💻 Quickstart
+
+### Installation
+
+```bash
+pip install numpy tokenizers safetensors huggingface_hub
+```
+
+### Python Inference
+
+```python
+from vortex_embed.src.lf4_v4_5 import VortexEmbedV4_5
+
+# Load 1M model from Hugging Face
+model = VortexEmbedV4_5.from_pretrained("VTXAI/vtx-embed-1M")
+
+sentences = ["Execute terminal command", "Run bash shell process in background"]
+embeddings = model.encode(sentences, normalize=True)  # (2, 64) float32
+
+sim = (embeddings[0] * embeddings[1]).sum()
+print(f"Similarity: {sim:.4f}")
+```
+
+---
+
+## 📜 Citation
+
+```bibtex
+@misc{vtx-embed-1M,
+  title  = {vtx-embed-1M: Data-Free Matrix Distillation for Sub-Megabyte Embeddings},
+  author = {VTXAI},
+  year   = {2026},
+  url    = {https://huggingface.co/VTXAI/vtx-embed-1M}
+}
+```
+
+---
+
+## 📄 License
+
+MIT License