{ "name": "Qwen3.5 Captioner", "description": "Applies a Qwen3.5 vision-language model to video frames selected by input TimeFrame annotations for prompt-driven captioning / scene description. Captioning is composite: the frames sampled from a TF are fed to the model in a single prompt and yield one caption. When `maxImagesPerPrompt` is set, a TF with more frames than the cap is split into several prompts and yields one caption per prompt, all aligned to that TF. A `model` runtime parameter selects the Qwen3.5 variant (default `Qwen/Qwen3.5-2B`); larger variants need substantially more VRAM (see metadata.py for the GPU/VRAM sketch). Qwen3.5 is a reasoning model -- this app runs it for captions and strips any `` reasoning block from the output.", "app_version": "v0.2", "mmif_version": "1.2.0", "analyzer_versions": { "Qwen/Qwen3.5-397B-A17B": "8472618", "Qwen/Qwen3.5-397B-A17B-FP8": "ea5b4f8", "Qwen/Qwen3.5-122B-A10B": "dc4d348", "Qwen/Qwen3.5-122B-A10B-FP8": "a099dee", "Qwen/Qwen3.5-35B-A3B": "59d61f3", "Qwen/Qwen3.5-35B-A3B-FP8": "9d1823d", "Qwen/Qwen3.5-27B": "fc05dae", "Qwen/Qwen3.5-27B-FP8": "97f5941", "Qwen/Qwen3.5-9B": "c202236", "Qwen/Qwen3.5-4B": "851bf6e", "Qwen/Qwen3.5-2B": "15852e8", "Qwen/Qwen3.5-0.8B": "2fc0636", "Qwen/Qwen3.5-397B-A17B-GPTQ-Int4": "df333de", "Qwen/Qwen3.5-122B-A10B-GPTQ-Int4": "30cd92c", "Qwen/Qwen3.5-35B-A3B-GPTQ-Int4": "3af5ca2", "Qwen/Qwen3.5-27B-GPTQ-Int4": "8f0c09f" }, "app_license": "Apache 2.0", "analyzer_license": "Apache 2.0", "identifier": "http://apps.clams.ai/qwen35-timeframe-captioner/v0.2", "url": "https://github.com/clamsproject/app-qwen35-timeframe-captioner", "input": [ { "@type": "http://clams.ai/vocabulary/type/VideoDocument/v2", "required": true }, { "@type": "http://clams.ai/vocabulary/type/TimeFrame/v6", "description": "Labeled TimeFrame annotations selecting which video segments to caption. Frame selection within each segment is controlled by the universal `tfSamplingMode` parameter (see SDK docs). Filter by label with the `tfLabels` parameter.", "properties": { "representatives": "?", "label": "*" }, "required": true } ], "output": [ { "@type": "http://clams.ai/vocabulary/type/TextDocument/v2", "description": "Caption text generated by the Qwen3.5 model for each prompt: one per processed TimeFrame, or several when `maxImagesPerPrompt` splits a TimeFrame. The `origins` property points to the `TimePoint`(s) anchoring the image(s) that caption was generated from.", "properties": { "origins": "*", "origination": "derived" } }, { "@type": "http://clams.ai/vocabulary/type/Alignment/v1", "description": "Alignment between each parent TimeFrame and the TextDocument(s) derived from it." }, { "@type": "http://clams.ai/vocabulary/type/TimePoint/v5", "description": "Optional output. Newly-created TimePoint annotations for images sampled from a TimeFrame interval without an existing backing TimePoint (see `tfSamplingMode`).", "properties": { "timeUnit": "milliseconds", "timePoint": "*" } } ], "parameters": [ { "name": "tfLabels", "description": "Label(s) of input TimeFrame annotations to caption. By default (`[]`), all TimeFrames are processed regardless of label.", "type": "string", "default": [], "multivalued": true }, { "name": "prompt", "description": "User prompt(s) sent to the model. A single value runs as a one-shot generation. A multi-value list is interpreted as a multi-turn static prompt; see ``promptMode`` for how turns are assembled.", "type": "string", "default": "These frames are sampled in order from one segment of a video. Describe what is shown and quote any visible text exactly. Describe only what the frames show; do not guess the date or era.", "multivalued": true }, { "name": "systemPrompt", "description": "Optional system-role text prepended to the conversation. Empty by default.", "type": "string", "default": "Answer in one plain-text paragraph. Do not use Markdown, headings, lists, or a separate summary.", "multivalued": false }, { "name": "promptMode", "description": "How to interpret a multi-value ``prompt`` list. Has no effect when ``prompt`` has a single value. For semantics of each choice and worked examples, see https://clams.ai/clams-python/app-baseclasses.html#promptable-multiturn", "type": "string", "choices": [ "user-only", "turn-taking" ], "default": "turn-taking", "multivalued": false }, { "name": "maxNewTokens", "description": "Maximum number of new tokens generated per inference call. Forwarded to the backend's ``generate``-equivalent. Larger values grow the KV cache linearly and increase GPU memory usage; reduce if VRAM is constrained.", "type": "integer", "default": 512, "multivalued": false }, { "name": "useReasoning", "description": "Request the model's reasoning (\"thinking\") mode. Off by default. Honored only by apps whose backing model has a distinct reasoning mode; apps without one ignore it. When honored and enabled, the reasoning trace is split from the answer and stored in the ``modelReasoningTrace`` property of the output ``TextDocument`` (kept out of the document text). Reasoning is markedly slower and far more token-hungry: the whole trace is generated before the answer and drawn from the same budget capped by ``maxNewTokens``, so raise ``maxNewTokens`` substantially (thousands of tokens, not hundreds) when enabling this, or the trace may consume the entire budget and the answer be truncated or empty. Small reasoning models (as a rule of thumb, roughly 4B parameters and under) are especially prone to non-terminating \"thinking loops\" that exhaust the budget without producing an answer; validate termination per model before relying on it.", "type": "boolean", "default": false, "multivalued": false }, { "name": "temperature", "description": "Sampling temperature. The default ``0.0`` selects deterministic / greedy decoding for maximum reproducibility; override for sampled generation.", "type": "number", "default": 0, "multivalued": false }, { "name": "topP", "description": "Nucleus-sampling cumulative probability cutoff. Only meaningful when ``temperature`` is greater than 0.", "type": "number", "default": 1, "multivalued": false }, { "name": "topK", "description": "Top-K sampling cutoff. Only meaningful when ``temperature`` is greater than 0.", "type": "integer", "default": 50, "multivalued": false }, { "name": "parallelPrompts", "description": "Number of independent prompts the app runs in parallel (stacks into a single forward pass). The *size* of each prompt (how many images, how long the system/user text is, etc.) is NOT regulated by this parameter; that is each app's responsibility. Prompt count and per-prompt content size combine multiplicatively for GPU memory, so the two can blow up together. Catastrophic example: ``tfSamplingMode=all`` on a TimeFrame without ``targets`` expands that TF into one image per native-FPS frame (300 images for a 10-second TF at 30fps); ``parallelPrompts=4`` then runs 4 such prompts in one forward pass (~1200 images), guaranteed OOM. Keep at ``1`` on memory-tight setups; raise only when per-prompt content is small and bounded.", "type": "integer", "default": 1, "multivalued": false }, { "name": "maxImagesPerPrompt", "description": "Maximum number of images the app bundles into a single prompt (one generation, one output TextDocument). ``0`` means no limit, so all images for a TimeFrame go into one prompt and yield one TextDocument. With a positive N, the app splits a TimeFrame whose frames exceed N into consecutive groups of at most N; each group produces its own TextDocument aligned to the same TimeFrame and grounded to its own source TimePoints via ``origins``. This bounds GPU memory for long dynamic scenes. The value is operator-set, so the number of output TextDocuments is deterministic and does not vary by GPU. Composes with ``parallelPrompts``, which batches independent prompts into one forward pass.", "type": "integer", "default": 0, "multivalued": false }, { "name": "imagesPerPromptMode", "description": "How the images of a TimeFrame that exceeds ``maxImagesPerPrompt`` are spread over its prompts. Has no effect when ``maxImagesPerPrompt`` is ``0``. Both choices produce the same number of prompts. \"max\" fills each prompt to the cap and puts the remainder in the last one (33 images at a cap of 32 give 32 and 1). \"balanced\" spreads the images evenly, so prompt sizes differ by at most one (33 images at a cap of 32 give 17 and 16).", "type": "string", "choices": [ "max", "balanced" ], "default": "max", "multivalued": false }, { "name": "model", "description": "HuggingFace model identifier to use for this request. Must be one of the model ids declared in this app's ``analyzer_versions``; the SDK pins the corresponding commit hash at load time. When the app ships a single model (the typical case), this parameter defaults to that one model and can be omitted. Pass the full HF model id (e.g. ``org/repo-name``); URL-encoding the ``/`` is optional.", "type": "string", "choices": [ "Qwen/Qwen3.5-397B-A17B", "Qwen/Qwen3.5-397B-A17B-FP8", "Qwen/Qwen3.5-122B-A10B", "Qwen/Qwen3.5-122B-A10B-FP8", "Qwen/Qwen3.5-35B-A3B", "Qwen/Qwen3.5-35B-A3B-FP8", "Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-27B-FP8", "Qwen/Qwen3.5-9B", "Qwen/Qwen3.5-4B", "Qwen/Qwen3.5-2B", "Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-397B-A17B-GPTQ-Int4", "Qwen/Qwen3.5-122B-A10B-GPTQ-Int4", "Qwen/Qwen3.5-35B-A3B-GPTQ-Int4", "Qwen/Qwen3.5-27B-GPTQ-Int4" ], "default": "Qwen/Qwen3.5-2B", "multivalued": false }, { "name": "pretty", "description": "The JSON body of the HTTP response will be re-formatted with 2-space indentation", "type": "boolean", "default": false, "multivalued": false }, { "name": "runningTime", "description": "The running time of the app will be recorded in the view metadata", "type": "boolean", "default": true, "multivalued": false }, { "name": "hwFetch", "description": "The hardware information (architecture, GPU and vRAM) will be recorded in the view metadata", "type": "boolean", "default": false, "multivalued": false }, { "name": "tfSamplingMode", "description": "Sampling mode for TimeFrame annotations. Has no effect when the app does not process TimeFrames. \"representatives\" uses all representative timepoints if present, otherwise skips the TimeFrame. \"single\" uses the middle representative if present, otherwise extracts an image from the midpoint of the start/end interval (midpoint is calculated by floor division of the sum of start and end). \"all\" uses all target timepoints if present, otherwise extracts all images from the time interval.", "type": "string", "choices": [ "representatives", "single", "all" ], "default": "representatives", "multivalued": false } ], "est_gpu_mem_min": 2000, "est_gpu_mem_typ": 5000 }