saathi-style-lora-r3 is an open-weight model from Krishna. Its published files total 819.5 MB.
Base model for the Lite tier of Saathi, a private offline journaling companion. - Converted with llama.cpp (b11192) from google/gemma-3-1b-it-qat-q40-unquantized.
Model Card
Base model for the Lite tier of Saathi, a private offline journaling companion. - Converted with llama.cpp (b11192) from google/gemma-3-1b-it-qat-q40-unquantized. Google's own q40 GGUF has slidingwindow 1024 and no freqbaseswa (the config says 512 / 10000), which breaks prompts longer than 512 tokens; this file fixes that. sha256 fe611e51b2e038f26bf77a1eb29bac385cbf1b1256629b6b255d444d2e8f590b - Perplexity (60 dialogue pairs, nctx 256): this file 49.60, Google's q40 49.34. Gemma is provided under and subject to the Gemma Terms of Use found at https://ai.google.dev/gemma/terms
Excerpt from the card by Krishna.
Identity and Version
- Repository
- krish99/saathi-style-lora-r3
- Publisher
- Krishna
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- Not stated by the source
- Languages
- Not stated by the source
- Revision
- f28cdb7713110fccc93fe9685badd5582f29e7e3
- First published
- 2026-09-27
- Last updated
- 2026-09-28
Files and Weights
6 files, 819.5 MB in total. The weights are 3 files totalling 819.5 MB in gguf.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| gemma-3-1b-it-qat-q4_0-saathi.gguf | Weights | 720.4 MB | fe611e51b2e0 |
| saathi-lora-r3-f16-qat.gguf | Weights | 48.3 MB | 7eb5d8d6e055 |
| saathi-lora-r3-f16.gguf | Weights | 50.7 MB | 17fc0714c366 |
| README-addition.md | Documentation | 658 B | — |
| README.md | Documentation | 707 B | — |
| .gitattributes | Repository | 1.7 KB | — |
License and Download
- License
- Not stated by the source
- Access
- Open weights, no gate
- Download size
- 819.5 MB
Released by Krishna through its official repository on Hugging Face.
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 819.5 MB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.