We have finished full fine-tuning of gemma-4-e2b-it for Nepali language.
This model was trained on 1xA100 GPU for approximately 9 days on 2M rows of mixed SFT dataset:
Nepali: himalaya-ai/nepali-sft-dataset English: teknium/OpenHermes-2.5.
We are open-sourcing the finetuned model himalaya-gemma-e2b-it on huggingface:
https://t.co/JpKl3KAZtD
@tridevgurung Me and team at @HimalayaAILabs are building Himmy Framework: agents with a real audit trail. Every tool call logged, costed, replayable, grounded in real tools, no hallucinating. First build: NEPSE Quant Desk, a terminal trading desk on local LLMs. Open-sourcing soon.
@Joshep777777 there's quiet nice pattern in #UHEWA on right side weekly yellow mean line act as great support area. pullback you can see on daily. left.