Transformers
Safetensors
llama
speculative-decoding
eagle3
draft-model
kimi-k2.5
fp8
amd-quark
quantized
no-lm-head-quantization
text-generation-inference
quark
Instructions to use amd/kimi-k2.5-eagle3-fp8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use amd/kimi-k2.5-eagle3-fp8 with Transformers:
# Load model directly from transformers import AutoTokenizer, LlamaForCausalLMEagle3 tokenizer = AutoTokenizer.from_pretrained("amd/kimi-k2.5-eagle3-fp8") model = LlamaForCausalLMEagle3.from_pretrained("amd/kimi-k2.5-eagle3-fp8") - Notebooks
- Google Colab
- Kaggle
| { | |
| "architectures": [ | |
| "LlamaForCausalLMEagle3" | |
| ], | |
| "attention_bias": false, | |
| "attention_dropout": 0.0, | |
| "bos_token_id": 163584, | |
| "draft_vocab_size": 163840, | |
| "dtype": "bfloat16", | |
| "eos_token_id": 163585, | |
| "head_dim": 128, | |
| "hidden_act": "silu", | |
| "hidden_size": 7168, | |
| "initializer_range": 0.02, | |
| "intermediate_size": 12288, | |
| "max_position_embeddings": 262144, | |
| "max_window_layers": 36, | |
| "mlp_bias": false, | |
| "model_type": "llama", | |
| "num_attention_heads": 64, | |
| "num_hidden_layers": 1, | |
| "num_key_value_heads": 64, | |
| "pretraining_tp": 1, | |
| "rms_norm_eps": 1e-06, | |
| "rope_scaling": null, | |
| "rope_theta": 1000000, | |
| "sliding_window": null, | |
| "tie_word_embeddings": false, | |
| "transformers_version": "4.57.1", | |
| "use_cache": true, | |
| "use_sliding_window": false, | |
| "vocab_size": 163840, | |
| "quantization_config": { | |
| "global_quant_config": { | |
| "input_tensors": { | |
| "dtype": "fp8_e4m3", | |
| "is_dynamic": true, | |
| "qscheme": "per_channel", | |
| "ch_axis": 1, | |
| "group_size": null, | |
| "block_size": null, | |
| "symmetric": true, | |
| "round_method": "half_even", | |
| "scale_type": "float", | |
| "scale_format": null, | |
| "scale_calculation_mode": null, | |
| "mx_element_dtype": null, | |
| "observer_cls": "PerChannelMinMaxObserver", | |
| "is_scale_quant": false, | |
| "enable_buffer_reuse": false, | |
| "max_input_numel": 4194304 | |
| }, | |
| "output_tensors": null, | |
| "weight": { | |
| "dtype": "fp8_e4m3", | |
| "is_dynamic": false, | |
| "qscheme": "per_channel", | |
| "ch_axis": 0, | |
| "group_size": null, | |
| "block_size": null, | |
| "symmetric": true, | |
| "round_method": "half_even", | |
| "scale_type": "float", | |
| "scale_format": null, | |
| "scale_calculation_mode": null, | |
| "mx_element_dtype": null, | |
| "observer_cls": "PerChannelMinMaxObserver", | |
| "is_scale_quant": false, | |
| "enable_buffer_reuse": false, | |
| "max_input_numel": 4194304 | |
| }, | |
| "bias": null, | |
| "target_device": null | |
| }, | |
| "exclude": [ | |
| "re:.*fc.*", | |
| "re:.*lm_head.*" | |
| ], | |
| "algo_config": null, | |
| "softmax_quant_spec": null, | |
| "quant_method": "quark", | |
| "layer_type_quant_config": {}, | |
| "layer_quant_config": {}, | |
| "kv_cache_quant_config": {}, | |
| "kv_cache_post_rope": false, | |
| "quant_mode": "eager_mode", | |
| "version": "0.12+5bd6865d5ca", | |
| "export": { | |
| "kv_cache_group": [], | |
| "min_kv_scale": 0.0, | |
| "pack_method": "reorder", | |
| "weight_format": "real_quantized", | |
| "weight_merge_groups": null | |
| } | |
| } | |
| } |