google-ai-edge/ai-edge-torch issues and pull requests

#363 - Memset prefill input tensors in text_generator_main binary.

Pull Request - State: closed - Opened by copybara-service[bot] about 21 hours ago

#362 - Convert PaliGemma.

Pull Request - State: open - Opened by copybara-service[bot] about 21 hours ago

#361 - Internal change. There is no other interesting things here.

Pull Request - State: open - Opened by copybara-service[bot] about 24 hours ago

#360 - Clean up BUILD files.

Pull Request - State: closed - Opened by copybara-service[bot] about 24 hours ago

#359 - models with upsampling layer can not be exported

Issue - State: open - Opened by bkunters 1 day ago - 2 comments
Labels: status:awaiting user response, type:bug

#358 - OPs list, allow experimental ops

Issue - State: open - Opened by gurkirt 1 day ago - 1 comment
Labels: status:awaiting user response, type:support

#357 - Fix the attribute name for odml.upsample_bilinear2d composite op.

Pull Request - State: open - Opened by copybara-service[bot] 3 days ago

#356 - Full-stack PaliGemma

Pull Request - State: closed - Opened by copybara-service[bot] 3 days ago

#355 - Failed to prepare model for QAT in Pt2e with quanconfig from ai_edge_torch

Issue - State: closed - Opened by gurkirt 4 days ago - 7 comments
Labels: status:awaiting user response, status:more data needed, type:bug, type:quantization

#354 - Add PaliGemma image encoder.

Pull Request - State: closed - Opened by copybara-service[bot] 4 days ago

#353 - Update max decode steps in text generator example to use KV max size.

Pull Request - State: closed - Opened by copybara-service[bot] 4 days ago

#352 - Create a guitar to catch numerical regression for gemma2 (torch). Exclude the guitar test from being copybara-ed.

Pull Request - State: open - Opened by copybara-service[bot] 4 days ago

#351 - Refactor verifier so that verification can return boolean value indicating if verification succeeds.

Pull Request - State: closed - Opened by copybara-service[bot] 4 days ago

#350 - Encounter 'normalizer.cc(52) LOG(INFO) precompiled_charsmap is empty. use identity normalization.' error when running exported models with AI Edge example.

Issue - State: open - Opened by merronmuche 5 days ago - 2 comments
Labels: status:awaiting user response, type:bug

#349 - Add is_nchw_op to the pattern of build_interpolate_composite_pass. Change existing pytorch composites to unify the upsample-bilinear composites from JAX and PyTorch.

Pull Request - State: closed - Opened by copybara-service[bot] 6 days ago

#348 - Raise on verification failures

Pull Request - State: closed - Opened by copybara-service[bot] 8 days ago

#347 - Enable HLFB in OpenELM

Pull Request - State: closed - Opened by copybara-service[bot] 8 days ago

#346 - Add PaliGemma decoder example to ai_edge_torch.

Pull Request - State: closed - Opened by copybara-service[bot] 8 days ago

#345 - Supporting LayerNorm Quantization

Issue - State: closed - Opened by ShinyCode 9 days ago - 3 comments
Labels: status:awaiting user response, type:support, type:quantization

#344 - Is there any plan support onnx model convertting to tflite ?

Issue - State: open - Opened by huanyingjun 10 days ago
Labels: type:feature, status:awaiting ai-edge-developer

#343 - remove tf-nightly version pin

Pull Request - State: closed - Opened by copybara-service[bot] 10 days ago

#342 - fix run_decomposition disrespecting tensor layout

Pull Request - State: closed - Opened by copybara-service[bot] 11 days ago

#341 - Don't enable hlfb for KV cache in OSS.

Pull Request - State: closed - Opened by copybara-service[bot] 11 days ago

#340 - Copybara scrub test

Pull Request - State: closed - Opened by copybara-service[bot] 11 days ago

#339 - Converting PyTorch Models to TensorFlow/TFLite with Training Support for Edge Devices

Issue - State: open - Opened by AdwaithAnand 12 days ago - 1 comment
Labels: type:feature, status:awaiting ai-edge-developer

#338 - Custom operator

Issue - State: open - Opened by fiberflow 13 days ago - 1 comment
Labels: type:feature, status:awaiting ai-edge-developer

#337 - Converting the fold operator resulted in an error: Only the Update, Add, Multiply, Maximum, and Minimum operations are currently supported for stablehlo.scatter. Node number 7 (STABLEHLO_SCATTER) failed to prepare

Issue - State: open - Opened by fiberflow 13 days ago - 1 comment
Labels: type:bug, status:awaiting ai-edge-developer

#336 - how to do inference in quantization model

Issue - State: open - Opened by shubhamgupto 14 days ago - 7 comments
Labels: status:awaiting user response, status:more data needed, type:bug, status:stale, type:quantization

#335 - Fix an issue in device type flag.

Pull Request - State: closed - Opened by copybara-service[bot] 14 days ago

#334 - set PJRT_DEVICE before loading torch_xla

Pull Request - State: closed - Opened by copybara-service[bot] 15 days ago

#333 - ai-edge-torch 0.2.1 release patches

Pull Request - State: closed - Opened by chunnienc 15 days ago
Labels: ci:run

#331 - Add direct lowering for aten.floor

Pull Request - State: closed - Opened by copybara-service[bot] 17 days ago

#330 - torch upsample convert error

Issue - State: open - Opened by dlgktjr 17 days ago - 1 comment
Labels: type:bug, status:awaiting ai-edge-developer

#329 - Update convert_to_tflite.py

Pull Request - State: open - Opened by haozha111 18 days ago

#328 - fix convert code.

Pull Request - State: closed - Opened by copybara-service[bot] 18 days ago

#327 - In SD decoder/diffusion configuration, add a device type field to check if hlfb is needed. Also update the converter code to inspect device type flag. This hopefully will limit the number of places where `enable_hlfb` should be flipped.

Pull Request - State: closed - Opened by copybara-service[bot] 18 days ago

#326 - Please document that PJRT_DEVICE=CPU is required

Issue - State: closed - Opened by artemisart 18 days ago - 5 comments
Labels: status:awaiting user response, type:support, status:stale

#325 - Making blocks_2d for SD U-net are more configurable

Pull Request - State: closed - Opened by alekseylebedevsc 19 days ago - 2 comments
Labels: ci:run, CLA:yes

#324 - Ensure all eKV buffers are `tflite::kDefaultTensorAlignment` aligned.

Pull Request - State: closed - Opened by copybara-service[bot] 19 days ago

#323 - Modify build dependencies for quantize example.

Pull Request - State: closed - Opened by copybara-service[bot] 19 days ago

#322 - Error while running Quantize example.py script for llama

Issue - State: open - Opened by Arya-Hari 21 days ago - 3 comments
Labels: type:bug, status:awaiting ai-edge-developer

#321 - Conversion keeps failing when trying to sub-sample a tensor using a list

Issue - State: closed - Opened by aneeshbhattacharya 22 days ago - 3 comments
Labels: status:awaiting user response, status:more data needed, type:bug, status:stale

#320 - Add a comment about multi-prefill export.

Pull Request - State: closed - Opened by copybara-service[bot] 22 days ago

#319 - internal change, not open sourced

Pull Request - State: open - Opened by copybara-service[bot] 22 days ago

#318 - Adds InferenceRunner class to execute the Stable Diffusion denoiser model via an InferenceCalculator instance.

Pull Request - State: open - Opened by copybara-service[bot] 22 days ago

#317 - nn.BatchNorm2d with affine=False

Issue - State: closed - Opened by shubhamgupto 23 days ago - 6 comments
Labels: status:awaiting user response, type:support

#316 - Fails to convert ConvTranspose2D with specific padding

Issue - State: open - Opened by alenhardt 23 days ago - 4 comments
Labels: type:bug, status:awaiting ai-edge-developer

#315 - Error while converting gemma2-2b

Issue - State: closed - Opened by Harshbande05 24 days ago - 5 comments
Labels: status:awaiting user response, type:support, status:stale

#314 - Error occurs when exporting tinyllama to tflite with num_layers=1

Issue - State: closed - Opened by hayyaw 24 days ago - 7 comments
Labels: status:awaiting user response, status:more data needed, status:stale

#313 - failed to convert 'SmolLM-135M' model to tflite. Error: 'tensorflow.lite.tools.flatbuffer_utils' has no attribute 'read_model_from_bytearray'

Issue - State: closed - Opened by merronmuche 25 days ago - 3 comments
Labels: status:awaiting user response, status:more data needed, type:bug, status:stale

#312 - Add AMD-Llama-135m to ai_edge_torch/generative/examples.

Pull Request - State: closed - Opened by copybara-service[bot] 25 days ago

#311 - Export multiple prefill seq lengths for Gemma2.

Pull Request - State: closed - Opened by copybara-service[bot] 25 days ago

#310 - odmltorch fix issues due to torchxla2 upgrade

Pull Request - State: closed - Opened by copybara-service[bot] 25 days ago

#309 - Fixes dependencies of stable_diffusion/samplers:init

Pull Request - State: closed - Opened by copybara-service[bot] 25 days ago

#308 - internal change, excluded from OSS

Pull Request - State: closed - Opened by copybara-service[bot] 25 days ago

#307 - Broken on Colab since torch-xla v2.5.0 was released

Issue - State: closed - Opened by kwikwag 27 days ago - 3 comments
Labels: status:awaiting user response, type:build/install, status:stale

#306 - Update tf-nightly requirement.

Pull Request - State: closed - Opened by copybara-service[bot] 29 days ago

#305 - failed to legalize operation 'tfl.pow' that was explicitly marked illegal

Issue - State: open - Opened by johndpope 29 days ago - 5 comments
Labels: type:bug, status:awaiting ai-edge-developer

#304 - Update README.md

Pull Request - State: closed - Opened by johndpope 29 days ago - 1 comment

#303 - Pin tf-nightly version to 1001 to avoid lowering to custom kv cache operation for generative api.

Pull Request - State: closed - Opened by copybara-service[bot] 29 days ago

#302 - upgrade requirements.txt deps to torch 2.5.0

Pull Request - State: open - Opened by copybara-service[bot] 30 days ago
Labels: ci:model-coverage

#301 - Dynamic Update Slice as a Torch Custom Op.

Pull Request - State: open - Opened by copybara-service[bot] about 1 month ago

#300 - Llama3 Conversion to TFLite version gets prematurely killed when running the convert_to_tflite.py script

Issue - State: open - Opened by Arya-Hari about 1 month ago - 5 comments
Labels: type:bug, status:awaiting ai-edge-developer

#299 - stabalize IR from torch export

Pull Request - State: closed - Opened by copybara-service[bot] about 1 month ago

#298 - Fix CanonicalizePass with contiguous reshape

Pull Request - State: closed - Opened by copybara-service[bot] about 1 month ago
Labels: ci:model-coverage

#297 - Use -inf for sliding window mask value

Pull Request - State: closed - Opened by copybara-service[bot] about 1 month ago

#296 - Request for tflite full int8 static quantization with calibration dataset

Issue - State: open - Opened by hayyaw about 1 month ago - 1 comment
Labels: type:feature, status:awaiting ai-edge-developer, type:quantization

#295 - Conversion fails on custom models (DCGAN, pix2pix)

Issue - State: open - Opened by jchwenger about 1 month ago - 3 comments
Labels: type:precision/accuracy, type:bug, status:awaiting ai-edge-developer

#294 - Update generative/README.md for tokenizer_to_sentencepiece.py

Pull Request - State: closed - Opened by copybara-service[bot] about 1 month ago

#293 - Error Using Converted Phi-3.5-mini TFLite in Android App

Issue - State: open - Opened by chienhuikuo about 1 month ago - 13 comments
Labels: type:feature, status:awaiting ai-edge-developer, type:performance

#292 - Add a pass to remove non user outputs from exported program

Pull Request - State: closed - Opened by copybara-service[bot] about 1 month ago

#291 - add jax bridge results elty sanitizer

Pull Request - State: open - Opened by copybara-service[bot] about 1 month ago
Labels: ci:model-coverage

#290 - add aten.cat direct lowering

Pull Request - State: open - Opened by copybara-service[bot] about 1 month ago
Labels: ci:model-coverage

#289 - Support Stable Diffusion on GPU

Pull Request - State: closed - Opened by Linchenn about 1 month ago

#288 - Torch `unsqueeze` / `reshape` dimensions not preserved upon TFLite conversion

Issue - State: open - Opened by gudgud96 about 1 month ago - 2 comments
Labels: type:bug, status:awaiting ai-edge-developer

#287 - Internal change only.

Pull Request - State: closed - Opened by copybara-service[bot] about 1 month ago

#286 - Getting an error while generating quantized tiny llama model due to AI-edge-quantizer

Issue - State: closed - Opened by pallaviNNT about 1 month ago - 5 comments
Labels: status:awaiting user response, status:more data needed, type:bug, status:stale

#285 - Remove use_input_shape option from LayerNorm since it should be False for all cases. Also updates GroupNorm and LayerNorm's default bias and weight as empty, referred to PyTorch implementation: https://pytorch.org/docs/stable/_modules/torch/nn/modules/normalization.html#LayerNorm, https://pytorch.org/docs/stable/_modules/torch/nn/modules/normalization.html#GroupNorm

Pull Request - State: closed - Opened by copybara-service[bot] about 1 month ago

GitHub / google-ai-edge/ai-edge-torch issues and pull requests