Talk it โ€” prompt KV cache state files

llama.cpp KV cache state files (llama_state_save_file) for the on-device expense parsing prompt of the Talk it app (Android). The app downloads the file that matches the user's settings so that the very first voice entry does not have to process the whole prompt from scratch (about 40 seconds on low-end phones without a GPU).

  • Model: Gemma 3 1B IT, Q4_K_M (ggml-org/gemma-3-1b-it-GGUF)
  • Generated on Android (CPU) with llama.rn 0.12.4 (llama.cpp build 9309, 6d57c26); n_ctx 2048, n_batch 32
  • Layout: <model id>/<hash>.kvstate. The hash is derived from the prompt and the engine settings, so a file is never overwritten. Old files are kept for older app versions.
  • Contents: only the app's fixed system prompt, fixed few-shot examples and the default category names. No user data.

Gemma Terms of Use

These files are computed by the Gemma model. Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms. Use is subject to the Gemma Prohibited Use Policy.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support