Sync model repo (text/metadata)

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,95 @@
+---
+library_name: executorch
+display_name: Gemma-4-E2B 8da4w — ExecuTorch
+license: apache-2.0
+tags:
+  - text-generation
+  - 8da4w
+  - quantized
+  - xnnpack
+  - kleidiai
+  - arm
+  - executorch
+  - edge-ai
+pipeline_tag: text-generation
+base_model: google/gemma-4-E2B
+base_model_relation: quantized
+---
+
+# Gemma-4-E2B 8da4w (ExecuTorch + XNNPACK)
+This is an **8da4w-quantized** (8-bit dynamic per-token activations + 4-bit per-channel grouped weights) version of [google/gemma-4-E2B](https://huggingface.co/google/gemma-4-E2B), optimized for edge deployment on ARM devices using [ExecuTorch](https://github.com/pytorch/executorch) with the **XNNPACK + KleidiAI** backend. 
+The model was quantized by ExecuTorch’s Gemma 4 export script without SDPA, with a custom static KV-cache, and exported to the `.pte` format for efficient on-device inference on ARM Cortex-A processors (modern Android phones, AWS Graviton, embedded ARM).
+
+## Key Highlights
+
+Compared to the FP32 baseline:
+
+- **2.6× smaller** — 10246.6 MB → 3000 MB on disk
+
+## Model Details
+
+### Model Description
+
+Gemma 4 E2B-IT is an instruction-tuned multimodal generative model from Google DeepMind. This text-only artifact is the same model after 8da4w post-training quantization and ExecuTorch/XNNPACK export, no fine-tuning was performed.
+Linear weights use groupwise INT4 quantization, activations are dynamically quantized to INT8, and embeddings use INT8.
+
+- **Developed by:** Google DeepMind (base model); Arm Model Optimization pipeline (quantization and export)
+- **Model type:** Text generation (autoregressive language model)
+- **License:** Apache 2.0
+- **Base model:** google/gemma-4-E2B-it (https://huggingface.co/google/gemma-4-E2B-it) — quantized and exported, not fine-tuned
+
+### Model Sources
+
+- **Repository:** https://huggingface.co/google/gemma-4-E2B
+- **Paper:** [Gemma 4 Technical Report](https://arxiv.org/abs/2607.02770)
+
+## How to Get Started with the Model
+
+## Evaluation
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+
+#### Metrics
+
+
+### Results
+
+## Technical Specifications
+
+### Objective
+
+Open-ended text generation and multiple-choice scoring with the Gemma 4 E2B architecture, deployed on ARM CPUs using INT4 linear weights and INT8 activations.
+
+### Quantization
+
+- **Activations**: 8-bit dynamic, per-token quantization
+- **Linear weights**: 4-bit, per-channel grouped quantization with group size 32
+- **Token embedding**: 8-bit, per-channel weight quantization
+- **Backend**: XNNPACK with KleidiAI INT4 micro-kernels
+- **Higher-precision layers**: None. Every linear layer and the token embedding are quantized. Attention internals remain in higher precision within the custom fused SDPA path.
+- **Calibration**: 200 prompts randomly sampled from the MMLU test split
+
+### Export Pipeline
+
+1. Load the pretrained Gemma 4 E2B-IT weights from Hugging Face safetensors and convert them into ExecuTorch’s custom Gemma 4 text-decoder architecture.
+2. Configure a static KV cache with a maximum sequence length of 1,024 tokens and enable dynamically sized input sequences.
+3. Replace standard attention with the fused llama::custom_sdpa tiled-attention operator.
+4. Apply TorchAO 8da4w post-training quantization:
+    - dynamic INT8 activations
+    - grouped INT4 linear weights
+    - group size 128
+    - incompatible layers retained at higher precision
+5. Quantize embedding tables to INT8. No calibration dataset or observer-calibration pass is used.
+6. Capture the quantized decoder with `torch.export` and decompose it into an ExecuTorch-compatible graph.
+7. Lower supported quantized and floating-point operations to the XNNPACK delegate and apply static memory planning.
+8. Serialize the graph and weights as an ExecuTorch `.pte` artifact compatible with the generic LLM runner.
+
+## Known Limitations
+
+- This is a text-only export. The original Gemma 4 E2B model’s image and audio encoders are not included.
+- 8da4w post-training quantization can reduce accuracy relative to the original model. Linear weights are INT4, dynamic activations are INT8, and embeddings are INT8.
+- Initial inference may incur one-time weight preparation or repacking overhead. Subsequent inference can be faster while the model remains loaded.
+

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f
+size 32169626