Instructions to use FastFlowLM/Gemma4-E2B-IT-NPU2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use FastFlowLM/Gemma4-E2B-IT-NPU2 with Transformers:
# Load model directly from transformers import AutoTokenizer, AutoModelForMultimodalLM tokenizer = AutoTokenizer.from_pretrained("FastFlowLM/Gemma4-E2B-IT-NPU2") model = AutoModelForMultimodalLM.from_pretrained("FastFlowLM/Gemma4-E2B-IT-NPU2") - Notebooks
- Google Colab
- Kaggle
Update config.json
Browse files- config.json +9 -4
config.json
CHANGED
|
@@ -42,24 +42,29 @@
|
|
| 42 |
],
|
| 43 |
"use_clipped_linears": true,
|
| 44 |
"Audio_MM_TILE_M": 128,
|
| 45 |
-
"Audio_MM_TILE_K":
|
| 46 |
"Audio_MM_TILE_N": 64,
|
| 47 |
"Gemma4E_Audio_audio_resample_rate": 16000,
|
| 48 |
"Gemma4E_Audio_gradient_clipping": 10000000000.0,
|
| 49 |
-
"
|
|
|
|
| 50 |
"Gemma4E_Audio_HIDDEN_SIZE": 1024,
|
| 51 |
-
"Gemma4E_Audio_INTERMEDIATE_SIZE":
|
| 52 |
"Gemma4E_Audio_attention_chunk_size": 12,
|
| 53 |
"Gemma4E_Audio_attention_context_left": 13,
|
| 54 |
"Gemma4E_Audio_attention_context_right": 0,
|
| 55 |
"Gemma4E_Audio_num_attention_heads": 8,
|
| 56 |
"Gemma4E_Audio_num_attention_layers": 12,
|
| 57 |
"Gemma4E_Audio_conv1d_kernel_size": 5,
|
|
|
|
| 58 |
"Gemma4E_conv2d_kernel_size": 3,
|
| 59 |
"Gemma4E_conv2d_Stride": 2,
|
| 60 |
"Gemma4e_conv2d_Padding": 1,
|
| 61 |
"Gemma4E_Audio_subsampling_conv_channels_0": 128,
|
| 62 |
-
"Gemma4E_Audio_subsampling_conv_channels_1": 32
|
|
|
|
|
|
|
|
|
|
| 63 |
},
|
| 64 |
"audio_token_id": 258881,
|
| 65 |
"boa_token_id": 256000,
|
|
|
|
| 42 |
],
|
| 43 |
"use_clipped_linears": true,
|
| 44 |
"Audio_MM_TILE_M": 128,
|
| 45 |
+
"Audio_MM_TILE_K": 256,
|
| 46 |
"Audio_MM_TILE_N": 64,
|
| 47 |
"Gemma4E_Audio_audio_resample_rate": 16000,
|
| 48 |
"Gemma4E_Audio_gradient_clipping": 10000000000.0,
|
| 49 |
+
"Gemma4E_Audio_Multimodal_Output_SIZE": 1536,
|
| 50 |
+
"Gemma4E_Audio_language_projection_output_size": 1536,
|
| 51 |
"Gemma4E_Audio_HIDDEN_SIZE": 1024,
|
| 52 |
+
"Gemma4E_Audio_INTERMEDIATE_SIZE": 4096,
|
| 53 |
"Gemma4E_Audio_attention_chunk_size": 12,
|
| 54 |
"Gemma4E_Audio_attention_context_left": 13,
|
| 55 |
"Gemma4E_Audio_attention_context_right": 0,
|
| 56 |
"Gemma4E_Audio_num_attention_heads": 8,
|
| 57 |
"Gemma4E_Audio_num_attention_layers": 12,
|
| 58 |
"Gemma4E_Audio_conv1d_kernel_size": 5,
|
| 59 |
+
"Gemma4E_Audio_conv1d_stride": 1,
|
| 60 |
"Gemma4E_conv2d_kernel_size": 3,
|
| 61 |
"Gemma4E_conv2d_Stride": 2,
|
| 62 |
"Gemma4e_conv2d_Padding": 1,
|
| 63 |
"Gemma4E_Audio_subsampling_conv_channels_0": 128,
|
| 64 |
+
"Gemma4E_Audio_subsampling_conv_channels_1": 32,
|
| 65 |
+
"Gemma4E_Audio_residual_weight": 0.5,
|
| 66 |
+
"Gemma4E_Audio_attention_softcap": 50.0
|
| 67 |
+
|
| 68 |
},
|
| 69 |
"audio_token_id": 258881,
|
| 70 |
"boa_token_id": 256000,
|