NEWVectors or files. Pick a path.Start →
    Served on MixpeekEmbeddingVerified 2026-11-05

    all-MiniLM-L6-v2

    sentence-transformers/all-MiniLM-L6-v2

    all-MiniLM-L6-v2 is a text embedding model (Apache-2.0). Served on Mixpeek, scaling to zero when idle.

    A Ray Serve app in Mixpeek's inference registry hosts it. It starts at zero replicas and wakes on first use.

    Use in a new namespaceAdd to an existing namespace Record as JSON

    New opens the Designer in Studio. For an existing namespace, open it in Studio and use its Designer tab.

    No demo for all-MiniLM-L6-v2 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
    Small, fast English sentence vectors on CPU (22M params, 384 dims) for short text, dedup and clustering.
    Worst at
    Non-English text, long passages (truncated at 256 tokens here), and hard retrieval where E5-large wins.
    Task
    Embedding
    Input
    Text
    Output
    Vector
    Use cases
    Text search

    Templates that use it

    No template uses this model yet. Browse templates.

    Input and output

    Input

    text
    max_tokens
    256
    truncation
    true
    language
    English
    prompts
    none

    Output

    vector
    dims
    384
    distance
    cosine
    type
    dense

    How Mixpeek wires it

    Extractors
    • text_extractor@v1 embedding_model=all_minilm_l6_v2_v1
    Feature URIs
    • mixpeek://text_extractor@v1/all_minilm_l6_v2_v1
    Vector indexes
    text_extractor_v1_embedding
    Retriever stages
    feature_search
    Combines with
    • text_extractor with embedding_model=all_minilm_l6_v2_v1 (dimensions fixed at namespace creation)

    How it runs

    Status
    Served on Mixpeek Ray Serve app in the inference registry; text_extractor can select it per run. The registry marks it a small fast fallback, E5 is the default.
    Serve apps
    sentence_transformers__all_MiniLM_L6_v2
    Runtime
    transformers AutoModel + AutoTokenizer (BertModel), mean pooling
    Weights
    safetensors, no remote code
    Parameters
    22.7 M
    Size on disk
    91 MB
    Preprocessing
    Tokenize with truncation at 256 tokens.
    Postprocessing
    Mean pool over the attention mask, L2-normalize (384 dims).
    Serving
    both
    Hardware
    Runs on CPU (slower without a GPU)

    Scale to zero

    Minimum replicas
    0
    Maximum replicas
    4
    Idle timeout
    300 s
    Memory
    0.5 GB
    CPUs
    0.25
    GPUs
    0
    Cold start, node warm
    Not recorded
    Cold start, new node
    Not recorded

    Measured cost

    Unmeasured

    No committed throughput or cost measurement for this model.

    Our evaluations

    Dataset and metricScoreBaselineDate
    LongMemEval subset, 60 questions (hybrid_flat, run outside Mixpeek)
    QA accuracy, SYSTEM number: local MiniLM dense + rank_bm25 RRF retrieval, Gemini reader; not Mixpeek's Serve app
    research/agent-memory/RESULTS.md
    		0.783None2026-09-27

    Questions

    What is all-MiniLM-L6-v2 best at?
    Small, fast English sentence vectors on CPU (22M params, 384 dims) for short text, dedup and clustering.
    What are the limits of all-MiniLM-L6-v2?
    Non-English text, long passages (truncated at 256 tokens here), and hard retrieval where E5-large wins.
    Can all-MiniLM-L6-v2 be used commercially?
    Yes. Apache-2.0 weights; no use restrictions.
    How does all-MiniLM-L6-v2 run on Mixpeek?
    Ray Serve app in the inference registry; text_extractor can select it per run. The registry marks it a small fast fallback, E5 is the default.