Instructions to use keras/qwen3_coder_instruct_30b_a3b_en with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- KerasHub
How to use keras/qwen3_coder_instruct_30b_a3b_en with KerasHub:
import keras_hub # Load CausalLM model (optional: use half precision for inference) causal_lm = keras_hub.models.CausalLM.from_preset("hf://keras/qwen3_coder_instruct_30b_a3b_en", dtype="bfloat16") causal_lm.compile(sampler="greedy") # (optional) specify a sampler # Generate text causal_lm.generate("Keras: deep learning for", max_length=64)import keras_hub # Create a Backbone model unspecialized for any task backbone = keras_hub.models.Backbone.from_preset("hf://keras/qwen3_coder_instruct_30b_a3b_en") - Keras
How to use keras/qwen3_coder_instruct_30b_a3b_en with Keras:
# !pip install -U keras tensorflow huggingface_hub # Keras needs TensorFlow installed to read "hf://" paths, so the tensorflow backend is selected here; # "jax" and "torch" also work for computation once TensorFlow is installed. import os os.environ["KERAS_BACKEND"] = "tensorflow" import keras model = keras.saving.load_model("hf://keras/qwen3_coder_instruct_30b_a3b_en") - Notebooks
- Google Colab
- Kaggle
Download preprocessor.json from keras/qwen3_coder_instruct_30b_a3b_en: direct link, hf CLI and curl.
- Browser
- Download file 2.42 kB
-
https://huggingface.co/keras/qwen3_coder_instruct_30b_a3b_en/resolve/main/preprocessor.json
- Command line
-
hf download hf://keras/qwen3_coder_instruct_30b_a3b_en/preprocessor.json
-
curl -L -o preprocessor.json https://huggingface.co/keras/qwen3_coder_instruct_30b_a3b_en/resolve/main/preprocessor.json
2.42 kB
| { | |
| "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_causal_lm_preprocessor", | |
| "class_name": "Qwen3MoeCausalLMPreprocessor", | |
| "config": { | |
| "name": "qwen3_moe_causal_lm_preprocessor_2", | |
| "trainable": true, | |
| "dtype": { | |
| "module": "keras", | |
| "class_name": "DTypePolicy", | |
| "config": { | |
| "name": "float32" | |
| }, | |
| "registered_name": null | |
| }, | |
| "tokenizer": { | |
| "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_tokenizer", | |
| "class_name": "Qwen3MoeTokenizer", | |
| "config": { | |
| "name": "qwen3_moe_tokenizer", | |
| "trainable": true, | |
| "dtype": { | |
| "module": "keras", | |
| "class_name": "DTypePolicy", | |
| "config": { | |
| "name": "int32" | |
| }, | |
| "registered_name": null | |
| }, | |
| "config_file": "tokenizer.json", | |
| "sequence_length": null, | |
| "add_prefix_space": false, | |
| "unsplittable_tokens": [ | |
| "<tool_response>", | |
| "</think>", | |
| "<|box_end|>", | |
| "</tool_response>", | |
| "<|fim_suffix|>", | |
| "<|quad_start|>", | |
| "<tool_call>", | |
| "<think>", | |
| "<|vision_pad|>", | |
| "<|file_sep|>", | |
| "<|box_start|>", | |
| "<|im_start|>", | |
| "<|endoftext|>", | |
| "</tool_call>", | |
| "<|object_ref_start|>", | |
| "<|repo_name|>", | |
| "<|vision_start|>", | |
| "<|vision_end|>", | |
| "<|object_ref_end|>", | |
| "<|quad_end|>", | |
| "<|fim_middle|>", | |
| "<|video_pad|>", | |
| "<|im_end|>", | |
| "<|image_pad|>", | |
| "<|fim_prefix|>", | |
| "<|fim_pad|>" | |
| ] | |
| }, | |
| "registered_name": "keras_hub>Qwen3MoeTokenizer" | |
| }, | |
| "config_file": "preprocessor.json", | |
| "sequence_length": 1024, | |
| "add_start_token": true, | |
| "add_end_token": true | |
| }, | |
| "registered_name": "keras_hub>Qwen3MoeCausalLMPreprocessor" | |
| } |