Repository navigation
Fix! - #50430
Fix!#50430
Conversation
|
[For maintainers] Suggested jobs to run (before merge) run-slow: phi4_multimodal, qianfan_ocr |
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
run-slow: phi4_multimodal, qianfan_ocr |
| audio_bytes = requests.get(self.audio_url, stream=True).raw.data | ||
| samples = torchcodec.decoders.AudioDecoder(audio_bytes).get_all_samples() | ||
| self.audio, self.sampling_rate = samples.data, samples.sample_rate | ||
| self.sampling_rate = 16000 | ||
| self.audio = load_audio(self.audio_url, sampling_rate=self.sampling_rate) |
There was a problem hiding this comment.
make_list_of_audio doesn't accept batched arrays (2D np/torch) and I dont want to "fix" it yet. IMO that is an ambiguous input - is that batched input or single multi-channel one. I believe it's the same issue with any model, not phi specific
Instead i fix it here to load using out utils
| ): | ||
| raise ValueError("`text` must be a string or list of strings.") | ||
|
|
||
| n_images_in_text = [sample.count(self.image_placeholder_token) for sample in text] |
There was a problem hiding this comment.
uses image_placeholder_token instead of image_token
There was a problem hiding this comment.
hmm maybe we could introduce an alias instead?
There was a problem hiding this comment.
they also have an image_token but used for different purposes 😭
CI (Nvidia + AMD)Nvidia · ✅ No new failures 🎉👏 · ⚙️ Workflow · 📊 Dashboard AMD · ❌ 3 tests with changed errors 😭 · ⚙️ Workflow · 📊 Dashboard Failures (3 pre-existing)Pre-existing failures (still failing) — 3 total: Error changed (3): - qianfan_ocr: |
| ): | ||
| raise ValueError("`text` must be a string or list of strings.") | ||
|
|
||
| n_images_in_text = [sample.count(self.image_placeholder_token) for sample in text] |
There was a problem hiding this comment.
hmm maybe we could introduce an alias instead?
as per title, fixes https://huggingface.co/datasets/hf-internal-testing/transformers_daily_ci/raw/5373d2a2164a11a055b6cc2ff1d8b19235fd37bc/2026-11-06/ci_results_run_models_gpu/new_failures_with_bad_commit_grouped_by_authors.json