Add GlucoFM encoder weights for Hugging Face publishing.

Safetensors F32, GGUF F16/Q8, upstream ONNX, model card, and checksums.

Co-authored-by: Cursor <cursoragent@cursor.com>

.gitattributes

@@ -33,3 +33,5 @@ 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
+*.gguf filter=lfs diff=lfs merge=lfs -text
+*.rlxp filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,4 @@
+.DS_Store
+*.tmp
+*.bak
+.cache/

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 GlucoFM contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

PUBLISH.md

@@ -0,0 +1,35 @@
+# Publish this folder as a Hugging Face model repo
+
+This directory is a **standalone** git repository. It is ignored by the parent
+`glucofm` code repo (see root `.gitignore`: `/artifacts/`).
+
+## One-time setup
+
+```bash
+cd artifacts
+# already initialized if you used the prep script; otherwise:
+#   git init
+#   git lfs install
+#   git lfs track "*.onnx" "*.safetensors" "*.gguf"
+
+huggingface-cli login
+# Create an empty model repo on HF (web UI or):
+#   huggingface-cli repo create glucofm-encoder --type model
+
+git remote add origin https://huggingface.co/<USER>/glucofm-encoder
+git add -A
+git commit -m "Add GlucoFM encoder weights (safetensors, GGUF, ONNX)"
+git push -u origin main
+```
+
+## Verify checksums
+
+```bash
+shasum -a 256 -c SHA256SUMS
+```
+
+## Suggested Hub id
+
+`glucofm-encoder` (or `<org>/glucofm-encoder`). After publish, point the
+parent crate README / `OPENCGM_HF_*` docs at that id if you replace the
+upstream OpenCGM download path.

README.md

@@ -0,0 +1,93 @@
+---
+license: mit
+library_name: glucofm
+tags:
+  - cgm
+  - glucose
+  - foundation-model
+  - jepa
+  - onnx
+  - safetensors
+  - gguf
+  - healthcare
+pipeline_tag: feature-extraction
+---
+
+# GlucoFM encoder weights
+
+Frozen **encoder-only** checkpoints for
+[GlucoFM](https://arxiv.org/abs/2605.30865) (dual-stream CGM foundation model),
+converted for the Rust [`glucofm`](https://crates.io/crates/glucofm) / RLX stack.
+
+> **Not medical advice.** Research reimplementation. Official Google GlucoFM
+> weights are not public. This package redistributes / converts a community
+> OpenCGM reconstruction.
+
+## Files
+
+| File | Format | Notes |
+|------|--------|-------|
+| `glucofm_encoder.safetensors` | F32 safetensors | **Preferred** for `glucofm` load / train warm-start |
+| `glucofm_encoder.f16.gguf` | GGUF F16 | Smaller disk; dequantized at load |
+| `glucofm_encoder.q8.gguf` | GGUF Q8_0 | ~4× smaller than F32; dequantized at load |
+| `glucofm_encoder.onnx` | ONNX opset 17 | Upstream OpenCGM export (raw grid → 128-d embedding) |
+| `glucofm_encoder.onnx.meta.json` | JSON | Upstream export metadata |
+| `config.json` | JSON | Geometry + param counts for this package |
+| `SHA256SUMS` | text | Integrity checks |
+
+**Encoder params:** 435,633 (includes `online/mask_token` + `online/gaussian.rho`).  
+OpenCGM ONNX omits `mask_token`; `glucofm` inference defaults it to zeros.
+
+Paper full trainable target (encoder + JEPA heads): **732,593**.
+
+## Geometry
+
+| Symbol | Value |
+|--------|------:|
+| Grid `L` | 288 (5 min) |
+| Patches `P` | 24 |
+| Patch len `K` | 12 |
+| Hidden `D` | 128 |
+| Stream `S` | 64 |
+
+## Provenance
+
+- Upstream community encoder: [`sfourdrinier/opencgm-stateevent`](https://huggingface.co/sfourdrinier/opencgm-stateevent)
+- OpenCGM project: [sfourdrinier/opencgm](https://github.com/sfourdrinier/opencgm)
+- Paper: [arXiv:2605.30865](https://arxiv.org/abs/2605.30865)
+- Conversion via `glucofm` `convert_weights` example (ONNX → safetensors / GGUF)
+
+Upstream ONNX meta (epoch 40 seed 17, raw stats, circadian, learnable σ) is kept in
+`glucofm_encoder.onnx.meta.json`.
+
+## Use with `glucofm`
+
+```bash
+# After cloning this HF repo next to your project, or downloading files:
+cargo run --release --features infer --example infer -- \
+  --weights /path/to/glucofm_encoder.safetensors
+
+cargo run --release --features hub --example train_from_hub -- \
+  --weights /path/to/glucofm_encoder.safetensors --steps 50
+```
+
+```rust,ignore
+use glucofm::prelude::*;
+
+let mut model = GlucoFm::for_inference(GlucoFmConfig::default(), InferOptions::default(), &mut rng)?;
+model.load_weights_path("glucofm_encoder.safetensors")?;
+let emb = model.encode_embedding_day(&day, &mut rng)?; // [128]
+```
+
+## Formats
+
+- Prefer **safetensors** for numeric fidelity with RLX `Param` graphs.
+- **GGUF** is for disk savings; runtime still uses dense F32/F16 params after load
+  (no in-graph Q8 matmul).
+
+## License
+
+Weights and packaging in this repository: **MIT** (see `LICENSE`).
+
+Upstream OpenCGM / paper licenses may impose additional terms on the original
+checkpoints — review those projects before commercial use.

SHA256SUMS

@@ -0,0 +1,5 @@
+glucofm_encoder.f16.gguf  7be81464df67fabb0e25b120a6de173c14aa2a346100c4b49a8a41b230305fcd
+glucofm_encoder.onnx  b1349deffd15ab62a5a98d7c7c4a7e143bcf1dee7ad745b1e05533ff54f34768
+glucofm_encoder.onnx.meta.json  a84d1870d526ff06a4a5a6fdd403e44b93db2c545e584138b6de0b23fcfe7773
+glucofm_encoder.q8.gguf  999e21cb0fa8ef9e6cf1386de11fcee796af827182353af537d26089b4788ce5
+glucofm_encoder.safetensors  8a9a3c471e09ddfcb35e076f80c0ef86520ab2f56172b1a01fa03c7edae22d3b

config.json

@@ -0,0 +1,36 @@
+{
+  "model_type": "glucofm",
+  "architectures": ["GlucoFmEncoder"],
+  "library_name": "glucofm",
+  "paper": "https://arxiv.org/abs/2605.30865",
+  "upstream_hub": "sfourdrinier/opencgm-stateevent",
+  "upstream_file": "glucofm_encoder.onnx",
+  "geometry": {
+    "grid_len": 288,
+    "num_patches": 24,
+    "patch_len": 12,
+    "hidden": 128,
+    "stream_dim": 64,
+    "num_heads": 4,
+    "ffn_dim": 256,
+    "num_layers": 3,
+    "delta_minutes": 5.0
+  },
+  "param_counts": {
+    "encoder": 435633,
+    "paper_trainable_full": 732593,
+    "note": "ONNX export omits online/mask_token (128); glucofm infer zeros it by default"
+  },
+  "files": {
+    "safetensors_f32": "glucofm_encoder.safetensors",
+    "gguf_f16": "glucofm_encoder.f16.gguf",
+    "gguf_q8": "glucofm_encoder.q8.gguf",
+    "onnx": "glucofm_encoder.onnx",
+    "onnx_meta": "glucofm_encoder.onnx.meta.json"
+  },
+  "io": {
+    "embedding_dim": 128,
+    "glucose_units": "mg/dL",
+    "mask_convention": "0 = unobserved, 1 = observed"
+  }
+}

glucofm_encoder.f16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7be81464df67fabb0e25b120a6de173c14aa2a346100c4b49a8a41b230305fcd
+size 876064

glucofm_encoder.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1349deffd15ab62a5a98d7c7c4a7e143bcf1dee7ad745b1e05533ff54f34768
+size 1991782

glucofm_encoder.onnx.meta.json

@@ -0,0 +1,38 @@
+{
+  "checkpoint": "runs_5090/rawstats120/ckpt_ep040.pt",
+  "epoch": 40,
+  "seed": 17,
+  "architecture": {
+    "raw_statistics": true,
+    "normalize_targets": false,
+    "streams": "both",
+    "learnable_sigma": true,
+    "use_circadian": true,
+    "zero_empty_patches": false
+  },
+  "opset": 17,
+  "input_names": [
+    "values",
+    "mask",
+    "circadian_start"
+  ],
+  "input_dtypes": [
+    "float32",
+    "float32",
+    "int64"
+  ],
+  "input_shapes": [
+    "[B, 288]",
+    "[B, 288]",
+    "[B]"
+  ],
+  "output_names": [
+    "embedding"
+  ],
+  "output_dtype": "float32",
+  "output_shape": "[B, 128]",
+  "size_bytes": 1991782,
+  "sha256": "b1349deffd15ab62a5a98d7c7c4a7e143bcf1dee7ad745b1e05533ff54f34768",
+  "mask_convention": "0 = unobserved, 1 = observed; never interpolate before this point",
+  "units": "mg/dL"
+}

glucofm_encoder.q8.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:999e21cb0fa8ef9e6cf1386de11fcee796af827182353af537d26089b4788ce5
+size 469360

glucofm_encoder.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a9a3c471e09ddfcb35e076f80c0ef86520ab2f56172b1a01fa03c7edae22d3b
+size 1749540