NEWVectors or files. Pick a path.Start →
    Runs as a custom pluginEmbeddingVerified 2026-11-05

    Qwen3 Embedding 0.6B

    Qwen/Qwen3-Embedding-0.6B

    Qwen3 Embedding 0.6B is a text embedding model (Apache-2.0). Can run on Mixpeek as a custom plugin.

    Mixpeek does not run it today. It can run as a custom plugin: the weights load without remote code.

    Record as JSON

    No demo for Qwen3 Embedding 0.6B yet

    A video, image, slideshow or playground goes here once we record one on our own data.

    What it is good and bad at

    Best at
    Instruction-aware multilingual text vectors from a small model (0.6B, 1024 dims) on CPU; in the Live Nation plugin it embeds Gemini tags plus the words spoken over each shot.
    Worst at
    Speech-only queries with one relevant shot (Live Nation speech category F1@10 0.249); it needs the Instruct/Query prompt on the query side, so a plain query embed underperforms.
    Task
    Embedding
    Input
    Text
    Output
    Vector
    Use cases
    Text search, Multilingual search, Transcripts

    Templates that use it

    No template uses this model yet. Browse templates.

    Input and output

    Input

    text
    max_tokens
    32768 (model limit)
    prompts
    queries: 'Instruct: <task> Query: <text>'; documents: no prefix

    Output

    vector
    dims
    1024
    distance
    cosine
    pooling
    last token
    normalized
    true

    How Mixpeek wires it

    Extractors
    • ln_shots@v1 fixed (customer _local plugin in customers/livenation/manifest-loop)
    Feature URIs
    • mixpeek://ln_shots@v1/ln_text
    Vector indexes
    ln_shots_v1_ln_text
    Retriever stages
    feature_search
    Combines with
    • ln_shots: whisper-large-v3-turbo speech + Gemini tags -> Qwen3 text vector, beside a SigLIP 2 so400m visual vector and BM25 on search_text, fused by DBSF

    How it runs

    Status
    Runs as a custom plugin Runs only inside the committed Live Nation _local plugin (ln_shots v1), loaded from /data/models as safetensors with remote code off; no Serve app or built-in extractor uses it.
    Serve apps
    None
    Runtime
    sentence_transformers.SentenceTransformer (Qwen3Model, last-token pooling, Normalize), device cpu
    Weights
    safetensors, no remote code
    Parameters
    596 M
    Size on disk
    1,192 MB
    Preprocessing
    Documents are the shot's tags and spoken sentences; queries get the Instruct/Query template with the Live Nation task.
    Postprocessing
    Last-token pooling and L2 normalization (sentence-transformers modules), batch 16.
    Serving
    both, batch 16
    Hardware
    Runs on CPU (slower without a GPU)

    Scale to zero

    Minimum replicas
    Not recorded
    Maximum replicas
    Not recorded
    Idle timeout
    Not recorded
    Memory
    Not recorded
    CPUs
    Not recorded
    GPUs
    Not recorded
    Cold start, node warm
    Not recorded
    Cold start, new node
    Not recorded

    Measured cost

    Unmeasured

    No committed throughput or cost measurement.

    Our evaluations

    Dataset and metricScoreBaselineDate
    Live Nation demo, 10 clips, 263 shots, 56 queries
    macro F1@10, variant 'text' (Qwen3 text vector alone, feature_search) (k=10)
    customers/livenation/manifest-loop/README.md
    		0.414the demo's native pipeline: 0.4122026-11-04
    Live Nation held-out clips, 28 queries the driver never saw
    macro F1@10, SYSTEM number: tri_dbsf_p200 baseline = SigLIP 2 so400m visual + Qwen3 text + BM25, DBSF fusion (k=10), held out
    customers/livenation/manifest-loop/DRIVER-RUN.md
    		0.406None2026-11-04

    Questions

    What is Qwen3 Embedding 0.6B best at?
    Instruction-aware multilingual text vectors from a small model (0.6B, 1024 dims) on CPU; in the Live Nation plugin it embeds Gemini tags plus the words spoken over each shot.
    What are the limits of Qwen3 Embedding 0.6B?
    Speech-only queries with one relevant shot (Live Nation speech category F1@10 0.249); it needs the Instruct/Query prompt on the query side, so a plain query embed underperforms.
    Can Qwen3 Embedding 0.6B be used commercially?
    Yes. Apache-2.0 weights; no use restrictions.
    How does Qwen3 Embedding 0.6B run on Mixpeek?
    Runs only inside the committed Live Nation _local plugin (ln_shots v1), loaded from /data/models as safetensors with remote code off; no Serve app or built-in extractor uses it.
    What does Qwen3 Embedding 0.6B output?
    A 1024-dimensional dense vector, compared by cosine similarity.