GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
jasonlee-sf
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
jasonlee-sf
Model Tree
Input Modalities
Output Modalities
Supported Functionality
nemotron_asr_pretrain_v1_10k_merged (NemotronH Nano Omni; hybrid Mamba/MoE LLM + sound encoder + vision).r=32, alpha=64, scaling 2.0) on q/k/v/o projections in the language-model attention layers [5, 12, 19, 26, 33, 42], plus a fully-trained sound_projection module.W += 2.0 · B @ A, accumulated in fp32 then cast to bf16) and the finetuned sound_projection.* tensors substituted for the base ones. All other weights are unchanged from the base model; keys/shapes/dtypes are identical, so no adapter is required at load time.python
from transformers import AutoModelForCausalLM, AutoProcessormodel = AutoModelForCausalLM.from_pretrained("jasonlee-sf/omni-slack-v28-15434-merged",trust_remote_code=True,torch_dtype="bfloat16",device_map="auto",)processor = AutoProcessor.from_pretrained("jasonlee-sf/omni-slack-v28-15434-merged", trust_remote_code=True)