Skip to content

Releases: mmm1712/Local_AI_esp32

PocketGPT v0.1.0 — deployment weights and tokenizer

Choose a tag to compare

@mmm1712 mmm1712 released this 14 Oct 03:18

Deployment weights and tokenizer for PocketGPT on ESP32-S3. Firmware is built from the repository source; these assets supply the previously missing model files.

  • DevKit: pocketgpt-devkit-assets-v0.1.0.zip — 32 MB Flash / 16 MB PSRAM, deduplicated Q4/FP16 pack plus tokenizer.
  • XIAO: pocketgpt-xiao-assets-v0.1.0.zip — 8 MB Flash / 8 MB PSRAM, Round Display and microSD; model partition, two caches and six SD runtime files.
  • SHA256SUMS.txt: archive integrity checks. Each ZIP also contains its own checksums, size/hash inventory, MIT model license and attribution notes.

Download, build and installation instructions

Use the package matching your board. DevKit and XIAO assets are not interchangeable. No prebuilt firmware, NVS/whole-board backup, credentials, datasets are included.

Verified offline: DevKit package hashes/footer/shared-matrix alias; XIAO 84 matrix placements and 168 digests, model/cache hashes and cache assembly; all 6,144 SentencePiece pieces/scores/types and normalizer match the original tokenizer after training-path metadata cleanup. All nine source tests pass.