Upload logs/convert_float16.log with huggingface_hub

logs/convert_float16.log

@@ -0,0 +1,73 @@
+[!] HF_TOKEN is not set. If model is public, you can pass --allow-no-token.
+[+] Using existing model dir: /home/ubuntu/translategemma-4b-it
+[+] ARCH: {'model_type': 'gemma3', 'architecture': 'Gemma3ForConditionalGeneration', 'vocab_size': 262208}
+[+] Strategy 1: litert-torch native
+[!] float16 support depends on converter/runtime build.
+[+] Gemma3 builders available: ['build_model_1b', 'build_model_270m']
+[+] Trying litert_torch.generative.examples.gemma3.gemma3.build_model_1b ...
+[!]   failed: 'NoneType' object has no attribute 'eval'
+[+] Trying litert_torch.generative.examples.gemma3.gemma3.build_model_270m ...
+[!]   failed: 'NoneType' object has no attribute 'eval'
+[!] Strategy 1 did not find a compatible builder
+[+] Strategy 2: ai_edge_torch generic (wrapped logits-only)
+[+] Loading HF model on CPU with dtype=torch.float16 ...
+`torch_dtype` is deprecated! Use `dtype` instead!
+
+Loading weights:   0%|                                                                                                                     | 0/883 [00:00<?, ?it/s]
+Loading weights:  15%|███████████████▋                                                                                          | 131/883 [00:00<00:00, 964.72it/s]
+Loading weights:  26%|███████████████████████████▎                                                                              | 228/883 [00:00<00:00, 818.54it/s]
+Loading weights:  35%|█████████████████████████████████████▏                                                                    | 310/883 [00:00<00:00, 656.66it/s]
+Loading weights:  43%|█████████████████████████████████████████████▍                                                            | 379/883 [00:00<00:00, 666.47it/s]
+Loading weights:  51%|██████████████████████████████████████████████████████                                                    | 450/883 [00:00<00:00, 679.41it/s]
+Loading weights:  92%|████████████████████████████████████████████████████████████████████████████████████████████████▊        | 814/883 [00:00<00:00, 1573.63it/s]
+Loading weights: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████| 883/883 [00:00<00:00, 1167.42it/s]
+(00:00) [START] LiteRT-Torch Convert
+(00:00) [START] LiteRT-Torch Convert > Torch Export: serving_default
+(00:08) [START] LiteRT-Torch Convert > Torch Export: serving_default > ExportedProgram Run Decompositions
+(00:27) [ DONE] LiteRT-Torch Convert > Torch Export: serving_default > ExportedProgram Run Decompositions (+00:18)
+(00:27) [ DONE] LiteRT-Torch Convert > Torch Export: serving_default (+00:27)
+(00:27) [START] LiteRT-Torch Convert > Run FX Passes
+(00:28) [START] LiteRT-Torch Convert > Run FX Passes > ExportedProgram Run Decompositions
+(00:28) [ DONE] LiteRT-Torch Convert > Run FX Passes > ExportedProgram Run Decompositions (+00:00)
+(00:28) [ DONE] LiteRT-Torch Convert > Run FX Passes (+00:01)
+(00:28) [START] LiteRT-Torch Convert > Lower to MLIR: serving_default
+(00:28) [START] LiteRT-Torch Convert > Lower to MLIR: serving_default > ExportedProgram Run Decompositions
+(00:46) [ DONE] LiteRT-Torch Convert > Lower to MLIR: serving_default > ExportedProgram Run Decompositions (+00:18)
+(00:46) [START] LiteRT-Torch Convert > Lower to MLIR: serving_default > ExportedProgram Run Decompositions
+(01:06) [ DONE] LiteRT-Torch Convert > Lower to MLIR: serving_default > ExportedProgram Run Decompositions (+00:19)
+(01:06) [START] LiteRT-Torch Convert > Lower to MLIR: serving_default > Create MLIR Module
+WARNING:jax._src.xla_bridge:An NVIDIA GPU may be present on this machine, but a CUDA-enabled jaxlib is not installed. Falling back to cpu.
+(01:30) [ DONE] LiteRT-Torch Convert > Lower to MLIR: serving_default > Create MLIR Module (+00:23)
+(01:30) [ DONE] LiteRT-Torch Convert > Lower to MLIR: serving_default (+01:01)
+(01:30) [START] LiteRT-Torch Convert > Merge MLIR Modules
+(01:30) [ DONE] LiteRT-Torch Convert > Merge MLIR Modules (+00:00)
+(01:30) [START] LiteRT-Torch Convert > Run LiteRT Converter Passes
+loc("__main__.strategy2_generic.<locals>.LogitsOnlyWrapper/transformers.models.gemma3.modeling_gemma3.Gemma3ForConditionalGeneration_model/transformers.models.gemma3.modeling_gemma3.Gemma3Model_model/transformers.models.gemma3.modeling_gemma3.Gemma3TextScaledWordEmbedding_embed_tokens;"("embedding"("/usr/local/lib/python3.12/dist-packages/torch/nn/modules/sparse.py":192:0))): error: failed to legalize operation 'tfl.embedding_lookup' that was explicitly marked illegal: %492 = "tfl.embedding_lookup"(%491, %274) : (tensor<128xi32>, tensor<262208x2560xf16>) -> tensor<128x2560xf16>
+(03:48) [ FAIL] LiteRT-Torch Convert > Run LiteRT Converter Passes
+(03:48) [ FAIL] LiteRT-Torch Convert
+[!] Strategy 2 failed: Failed to run converter passes: /usr/local/lib/python3.12/dist-packages/torch/nn/modules/sparse.py:192:0: error: failed to legalize operation 'tfl.embedding_lookup' that was explicitly marked illegal: %492 = "tfl.embedding_lookup"(%491, %274) : (tensor<128xi32>, tensor<262208x2560xf16>) -> tensor<128x2560xf16>
+
+Traceback (most recent call last):
+  File "/home/ubuntu/convert_translategemma_android.py", line 372, in main
+    tflite_file = strategy2_generic(
+                  ^^^^^^^^^^^^^^^^^^
+  File "/home/ubuntu/convert_translategemma_android.py", line 234, in strategy2_generic
+    edge_model = ai_edge_torch.convert(wrapped, (sample_ids,))
+                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/litert_torch/_convert/interface.py", line 323, in convert
+    return Converter().convert(
+           ^^^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/litert_torch/_convert/interface.py", line 208, in convert
+    converted_model = core.convert_signatures(
+                      ^^^^^^^^^^^^^^^^^^^^^^^^
+  File "/usr/lib/python3.12/contextlib.py", line 81, in inner
+    return func(*args, **kwds)
+           ^^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/litert_torch/_convert/core.py", line 158, in convert_signatures
+    exporter = litert_converter.exported_programs_to_flatbuffer(
+               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+  File "/usr/local/lib/python3.12/dist-packages/litert_torch/_convert/litert_converter.py", line 189, in exported_programs_to_flatbuffer
+    converter_api_ext.run_convert_to_tfl_passes(
+ValueError: Failed to run converter passes: /usr/local/lib/python3.12/dist-packages/torch/nn/modules/sparse.py:192:0: error: failed to legalize operation 'tfl.embedding_lookup' that was explicitly marked illegal: %492 = "tfl.embedding_lookup"(%491, %274) : (tensor<128xi32>, tensor<262208x2560xf16>) -> tensor<128x2560xf16>
+
+[x] All conversion strategies failed