| ---
|
| pipeline_tag: text-generation
|
| license: apache-2.0
|
| base_model: Qwen/Qwen2-7B-Instruct
|
| library_name: kerasformers
|
| language:
|
| - en
|
| tags:
|
| - keras
|
| - kerasformers
|
| - qwen2
|
| - text-generation
|
| - pytorch
|
| - jax
|
| - tf
|
| ---
|
|
|
| Paper: [Qwen2 Technical Report (arXiv:2407.10671)](https://arxiv.org/abs/2407.10671) · [HF Papers](https://huggingface.co/papers/2407.10671)
|
|
|
| ## ***See [our collection](https://huggingface.co/collections/kerasformers/qwen2-6a69d274d16370be5d0221c8) for all Qwen2 versions.***
|
|
|
| # Run Qwen2 with Keras 3: JAX, PyTorch, or TensorFlow
|
|
|
| [](https://github.com/IMvision12/KerasFormers) [](https://imvision12.github.io/KerasFormers/qwen2/) [](https://huggingface.co/collections/kerasformers/qwen2-6a69d274d16370be5d0221c8)
|
|
|
| # kerasformers/qwen2-7b-instruct
|
|
|
| Qwen2 is Alibaba's decoder-only transformer family: grouped-query attention with q/k/v bias, SwiGLU MLPs, RMSNorm, and rotary positions, in dense 0.5B-72B sizes (plus the Qwen2-57B-A14B mixture-of-experts), as base and instruct variants.
|
|
|
| For more details on the model, please see the upstream [model card](https://huggingface.co/Qwen/Qwen2-7B-Instruct).
|
|
|
| Pure-**Keras 3** conversion of [`Qwen/Qwen2-7B-Instruct`](https://huggingface.co/Qwen/Qwen2-7B-Instruct) for [kerasformers](https://github.com/IMvision12/KerasFormers). One implementation runs unmodified on **TensorFlow / Torch / JAX**.
|
|
|
| This is an **instruct** (chat-tuned) checkpoint; load `Qwen2Tokenizer` so the chat template is applied.
|
|
|
| ## Quick start
|
|
|
| ```python
|
| import os
|
| os.environ["KERAS_BACKEND"] = "torch" # or "jax" / "tensorflow"
|
|
|
| from kerasformers.models.qwen2 import Qwen2TextGenerate, Qwen2Tokenizer
|
|
|
| model = Qwen2TextGenerate.from_weights("kerasformers/qwen2-7b-instruct")
|
| tokenizer = Qwen2Tokenizer.from_weights("kerasformers/qwen2-7b-instruct")
|
|
|
| inputs = tokenizer([
|
| {"role": "user", "content": "Explain rotary embeddings in one sentence."}
|
| ])
|
| outputs = model.generate(**inputs, max_new_tokens=64)
|
| print(tokenizer.decode(outputs[0]))
|
| ```
|
|
|
| Load any Qwen2 variant the same way with `from_weights("kerasformers/<variant>")`:
|
|
|
| | Variant | Hub | Type |
|
| |---|---|---|
|
| | `qwen2-0.5b` | [`kerasformers/qwen2-0.5b`](https://huggingface.co/kerasformers/qwen2-0.5b) | base |
|
| | `qwen2-0.5b-instruct` | [`kerasformers/qwen2-0.5b-instruct`](https://huggingface.co/kerasformers/qwen2-0.5b-instruct) | instruct |
|
| | `qwen2-1.5b` | [`kerasformers/qwen2-1.5b`](https://huggingface.co/kerasformers/qwen2-1.5b) | base |
|
| | `qwen2-1.5b-instruct` | [`kerasformers/qwen2-1.5b-instruct`](https://huggingface.co/kerasformers/qwen2-1.5b-instruct) | instruct |
|
| | `qwen2-7b` | [`kerasformers/qwen2-7b`](https://huggingface.co/kerasformers/qwen2-7b) | base |
|
| | `qwen2-7b-instruct` | [`kerasformers/qwen2-7b-instruct`](https://huggingface.co/kerasformers/qwen2-7b-instruct) | instruct |
|
| | `qwen2-72b` | [`kerasformers/qwen2-72b`](https://huggingface.co/kerasformers/qwen2-72b) | base |
|
| | `qwen2-72b-instruct` | [`kerasformers/qwen2-72b-instruct`](https://huggingface.co/kerasformers/qwen2-72b-instruct) | instruct |
|
| | `qwen2-57b-a14b` | [`kerasformers/qwen2-57b-a14b`](https://huggingface.co/kerasformers/qwen2-57b-a14b) | MoE base |
|
| | `qwen2-57b-a14b-instruct` | [`kerasformers/qwen2-57b-a14b-instruct`](https://huggingface.co/kerasformers/qwen2-57b-a14b-instruct) | MoE instruct |
|
| | `qwen1.5-moe-a2.7b` | [`kerasformers/qwen1.5-moe-a2.7b`](https://huggingface.co/kerasformers/qwen1.5-moe-a2.7b) | MoE base |
|
| | `qwen1.5-moe-a2.7b-chat` | [`kerasformers/qwen1.5-moe-a2.7b-chat`](https://huggingface.co/kerasformers/qwen1.5-moe-a2.7b-chat) | MoE chat |
|
|
|
| ## Tips
|
|
|
| - Set `KERAS_BACKEND` **before** importing Keras / kerasformers.
|
| - Prefer `Qwen2Tokenizer.from_weights(...)` so the chat template matches.
|
| - Larger checkpoints: try `load_dtype="bfloat16"` or `quantization="int8"`.
|
| - See [Loading Weights](https://imvision12.github.io/KerasFormers/loading_weights/) and the [Qwen2 docs](https://imvision12.github.io/KerasFormers/qwen2/).
|
| - Community / upstream safetensors still work via the `hf:` prefix, e.g. `Qwen2TextGenerate.from_weights("hf:Qwen/Qwen2-7B-Instruct")`.
|
|
|
| ## Special Thanks
|
|
|
| A huge thank you to the Qwen team at Alibaba for creating and releasing these models.
|
|
|
| License: Apache 2.0.
|
|
|
