Ma77Ball commented on code in PR #5320: URL: https://github.com/apache/texera/pull/5320#discussion_r3352581349
########## common/workflow-operator/src/main/scala/org/apache/texera/amber/operator/huggingFace/codegen/ImageTaskCodegen.scala: ########## @@ -0,0 +1,146 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +package org.apache.texera.amber.operator.huggingFace.codegen + +/** + * Codegen for the Hugging Face image-pipeline task family. + * + * Splits into two sub-families: + * - "image-only" tasks send raw image bytes as the request body and don't + * consume the prompt column: image-classification, object-detection, + * image-segmentation, image-to-text. + * - "image + prompt" tasks bundle a base64 image and a text prompt in a + * JSON payload: visual-question-answering, document-question-answering, + * zero-shot-image-classification, image-text-to-text, image-to-image. + * + * Per-row `current_image_bytes` is resolved upstream in + * [[PythonCodegenBase]]'s `process_table` (either from the operator's + * uploaded image or from `INPUT_IMAGE_COLUMN`). The image helpers + * (`_read_image_input`, `_compress_image_bytes`, `_image_input_as_base64`, + * `_read_binary_value`, `_looks_like_html`, `_html_to_image_bytes`, + * `_extract_json_arg`) live in PythonCodegenBase alongside the per-task + * tuples (`image_only_tasks`, `image_prompt_tasks`, `image_tasks`). + */ +object ImageTaskCodegen extends TaskCodegen { + + /** Primary key for registration; the dispatcher maps every task in + * [[tasks]] to this codegen. + */ + override val task: String = "image-classification" + + /** All HF tasks routed through this codegen. */ + override val tasks: Set[String] = Set( + // image-only + "image-classification", + "object-detection", + "image-segmentation", + "image-to-text", + // image + prompt + "visual-question-answering", + "document-question-answering", + "zero-shot-image-classification", + "image-text-to-text", + "image-to-image" + ) + + override def payloadPython(ctx: CodegenContext): String = + """ if task in image_only_tasks: + | payload = current_image_bytes + | use_raw_binary_body = True + | raw_binary_headers = image_headers + | elif task in ("visual-question-answering", "document-question-answering"): + | payload = { + | "inputs": { + | "image": self._image_input_as_base64(current_image_bytes), + | "question": prompt_value, + | } + | } + | elif task == "image-text-to-text": + | img_b64 = self._image_input_as_base64(current_image_bytes) + | payload = { + | "model": self.MODEL_ID, + | "messages": [{ + | "role": "user", + | "content": [ + | {"type": "image_url", "image_url": {"url": f"data:image/png;base64,{img_b64}"}}, + | {"type": "text", "text": prompt_value if prompt_value else "Describe this image."}, + | ], + | }], + | "max_tokens": self.MAX_NEW_TOKENS, + | } + | elif task == "image-to-image": + | payload = current_image_bytes + | use_raw_binary_body = True + | raw_binary_headers = image_headers + | elif task == "zero-shot-image-classification": + | labels = [] Review Comment: `candidate_labels` is hardcoded to `[]`, so `zero-shot-image-classification` always sends an empty label set and the API cannot return a useful prediction: the task is effectively dead. Zero-shot classification only works when the caller supplies the candidate labels. Why this suggestion is safe to apply: zero-shot is already in `image_prompt_tasks`, so `prompt_value` is populated for this row, and line 96 already reads `labels`, so only this line needs to change. Reusing the prompt column as a comma-separated label list makes the task functional with no new fields: ```suggestion | labels = [s.strip() for s in prompt_value.split(",") if s.strip()] ``` Longer term, a dedicated `candidateLabels` operator field is cleaner than overloading the prompt column, but this keeps the task shippable now instead of advertising one that cannot run. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
