Add tiny random LFM3-Omni model for optimum-intel tests
#1
by rkazants - opened
Random 2.90M-parameter LFM3-Omni (remote code) with a 1.91M-parameter Liquid Codec in audio_tokenizer/. The language model keeps every layer type (conv, full and sliding attention, sliding window 4). The vision tower, speech encoder, depth decoder and codec are all shrunk. Built with make_tiny_lfm3_omni.py from LiquidAI/LFM3-Omni-2.2B.
rkazants changed pull request status to open
rkazants changed pull request status to merged