Downloads · 30 days
131
38% of all-time downloads
ganmoor-ai-labs/krishi-sathi-sarvam-2b-multilingual
krishi-sathi-sarvam-2b-multilingual is a text generation model from ganmoor-ai-labs. Use it when you need the model to write or continue text. The card lists the license as other.
⚠️ Disclaimer — experimental / educational use only Krishi Sathi is an experimental research and education project, provided as-is with no warranty of any kind. Large language models can produce incorrect or outdated…
Downloads · 30 days
131
38% of all-time downloads
All-time downloads
348
Public
Repo size
23.5 GB
Likes
0
Public
Click a slice to open those files.
.gguf6.6 GB · 56%
From the Hugging Face model README
⚠️ Disclaimer — experimental / educational use only
Krishi Sathi is an experimental research and education project, provided as-is with no warranty of any kind. Large language models can produce incorrect or outdated information. This model is not professional agricultural advice; the authors accept no responsibility or liability for decisions made or losses incurred based on its outputs. Always verify any chemical, dosage, variety or practice with the product label and your local agriculture office (Raitha Samparka Kendra / Rythu Vedika / RBK) before acting.
Krishi Sathi ("farmer's companion") is a small trilingual farmer-advisory model built to run fully offline on low-end Android phones. This release extends the Kannada-only flagship with Telugu and Tamil, for farmers in Karnataka, Telangana, Andhra Pradesh and Tamil Nadu.
Like many 2B-class Indic models, this model can fall into repetition loops under plain sampling (~15–27% of long generations in our benchmark). Serve it with llama.cpp's DRY sampler, which eliminated loops entirely (0/80) in our benchmark with no measurable quality cost:
llama-server -m krishi-sathi-sarvam-2b-multilingual-q4_k_m.gguf \
-c 4096 -ngl 99 --jinja \
--dry-multiplier 0.8 --dry-sequence-breaker "\n" --dry-penalty-last-n 512
Note: the default DRY sequence breakers include *, which makes DRY blind to
this model's characteristic * * * loops — overriding the breakers as above is
required, not optional. Ollama does not currently expose DRY, which is why this
release is llama.cpp-first.
Recommended sampling: temperature 0.3, top_p 0.9, repeat_penalty 1.1,
max ~700 tokens, stop on [INST].
dosage_table.json (985 label-derived
rows, 98 crops) is provided for app-side grounding of dosages.Inherits the Sarvam AI Research License from sarvam-1: non-commercial research and evaluation use only. See LICENSE.md in this repo. Redistribution of derivatives must include the license and attribution.
Built by Ganmoor AI Labs. Kisan Call Centre data via
Omegaindebt/Kisan_Call_Centre_Transcripts; translations via IndicTrans2
(prajdabre/rotary-indictrans2-en-indic-1B).