GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
didiudom94
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
didiudom94
Model Tree
openai/whisper-small
Input Modalities
Output Modalities
Supported Functionality
task="translate")openai/whisper-smallNF4) via bitsandbytesThe model was trained on a robust 66-hour master dataset fusion matrix, balancing domain-specific variety show vocabulary with general conversational stability using a precise 70:30 ratio:
didiudom94/my-zeroth-audio-dataset-with-text (4,000 balanced rows)Bingsu/KSS_Dataset (10,000 balanced rows)["q_proj", "v_proj", "out_proj", "k_proj"])per_device_train_batch_size=32, gradient_accumulation_steps=1)bfloat16 (bf16=True, tf32=True)metric_for_best_model="loss")Evaluation was run using a 1,000-row unseen evaluation slice drawn directly from the master test matrix. The model demonstrates a near-doubling of string-level contextual translation accuracy over the original out-of-the-box Whisper Small base weights.
| Model Variant | Corpus-Level BLEU Score | Translation Alignment Behavior |
|---|---|---|
| 📉 Original Base OpenAI Whisper Small | 5.55 | Struggles with colloquial variety filters; outputs pure Korean transcription or severe string hallucinations. |
| 🚀 Fine-Tuned Burn-to-Win V2 (This Model) | 10.68 | Accurately interprets sports enthusiasm and intent, generating fluent human-like subbing choices. |
how is he that fast ?why is he so fast ? (Highly contextual, fluent substitution)we should find nobody (Complete hallucination)To load this fine-tuned checkpoint directly from the Hugging Face Hub using the peft and transformers ecosystem, use the following code snippet:
python
import torchfrom transformers import WhisperProcessor, WhisperForConditionalGenerationfrom peft import PeftModel, PeftConfig peft_model_id = "YOUR_HF_USERNAME/YOUR_REPO_NAME" # 1. Load base configuration and processorconfig = PeftConfig.from_pretrained(peft_model_id)processor = WhisperProcessor.from_pretrained(config.base_model_name_or_path, language="english", task="translate") # 2. Initialize underlying base architecture base_model = WhisperForConditionalGeneration.from_pretrained( config.base_model_name_or_path, torch_dtype=torch.bfloat16, attn_implementation="sdpa", device_map="auto") # 3. Inject fine-tuned adapter layersmodel = PeftModel.from_pretrained(base_model, peft_model_id)model.eval()print("🎉 Custom translation engine deployed successfully!")stable-ts (Stable Whisper) with arguments chunk_size=30 and condition_on_previous_text=False to maintain constant timestamp stability and eliminate trailing token duplication bugs.import torchfrom transformers import WhisperProcessor, WhisperForConditionalGenerationfrom peft import PeftModel, PeftConfig peft_model_id = "YOUR_HF_USERNAME/YOUR_REPO_NAME" # 1. Load base configuration and processorconfig = PeftConfig.from_pretrained(peft_model_id)processor = WhisperProcessor.from_pretrained(config.base_model_name_or_path, language="english", task="translate") # 2. Initialize underlying base architecture base_model = WhisperForConditionalGeneration.from_pretrained( config.base_model_name_or_path, torch_dtype=torch.bfloat16, attn_implementation="sdpa", device_map="auto") # 3. Inject fine-tuned adapter layersmodel = PeftModel.from_pretrained(base_model, peft_model_id)model.eval()print("🎉 Custom translation engine deployed successfully!")