test_model_snowflake_l/README.md

metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:5776
  - loss:MultipleNegativesRankingLoss
base_model: Snowflake/snowflake-arctic-embed-l
widget:
  - source_sentence: >-
      What is the primary function of the "EFFECTOR" as described in the
      document titled "EFFECTOR"?
    sentences:
      - >-
        Document_title: Sawer - Modulation

        File_name: plugins/Sawer_Modulation.htm

        Heading_hierarchy: [Sawer - Modulation -> LFO Controls]

        Anchor_id: [none]

        LFO is a 'Low Frequency Oscillator' that creates an oscillating control
        output, the rate of which can be adjusted between 0 and 10 Hz. This
        generator is used to control parameters inside Sawer to add motion
              or variability to sounds in order to make them more interesting. • SPEED - LFO speed, between 0 and 10 Hz. • ATTACK - Fade-in for the LFO, ramps up the LFO amount. This is linked to a note-on event. • RELEASE - Fade-out for the LFO, ramps down the LFO amount. This is linked to a
        note-release event. • AMPLITUDE - A multiplier or amount of LFO to be
        applied to the Destination control. • RETRIG (Retrigger) - Restarts the
        LFO cycle at the start of each note. • TMP SYNC (Tempo Sync) -
        Synchronizes the LFO speed to the host tempo. When selected the SPEED
        knob now

        controls the tempo sync from 1/16th to 32/4. • INV (Invert) - Inverts
        the phase of the LFO waveform. • SHAPE - Select the LFO shape (TRIANGLE,
        SAW, SQUARE, RANDOM). • LFO DESTINATION LFO Destination, select from - ➤
        CUTOFF - [Filter cutoff

        frequency](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Sawer_Filter.htm)
        . ➤ SYNC FRQ (Frequency) - [Main oscillator synchronization
        frequency](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Sawer_Oscillators.htm)
        . ➤ PHASE - [Sub-o

        scillator
        phase](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Sawer_Oscillators.htm)
        . ➤ FREQ (Frequency) - [Master tuning
        frequency](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Sawer_Master.htm)
        . ➤ NOISE - Controls the [Oscillato

        r noise
        level](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Sawer_Oscillators.htm)
        . ➤ LEVEL - [Master
        volume](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Sawer_Master.htm)
        . ➤ PAN - [Master

        panning](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Sawer_Master.htm)
        .
      - |-
        Document_title: EFFECTOR
        File_name: plugins/Effector.htm
        Heading_hierarchy: [EFFECTOR -> Controls]
        Anchor_id: [none]
        From top to bottom, left to right:
      - >-
        Document_title: FL Keys

        File_name: plugins/FL Keys.htm

        Heading_hierarchy: [FL Keys]

        Anchor_id: [none]

        FL Keys With the help of several additional modules included in FL
        Studio, FL Keys can also operate as Rhodes & Roto Organ.
  - source_sentence: >-
      How does GDPR privacy regulation affect the process of regaining access to
      an FL Studio account if a user loses access and their identity cannot be
      verified?
    sentences:
      - >-
        Title: Legal name change

        Answer: If you are looking to hide your name to protect your online
        privacy when using FL Studio in public, see - [How can I change the name
        displayed in the Title
        Bar?](https://support.image-line.com/action/knowledgebase/?ans=524)To
        change the name on your Account after a legal name change: Please create
        a [Support Ticket
        here](https://support.image-line.com/action/ticketing/add-issue), attach
        your documentation and request a name change for your license or
        account. NOTE: Always ensure you have a verifiable user ID in your
        account. With FL Studio's Lifetime Free Updates,

        it's crucial for us to identify you as the rightful owner. If you lose
        access to your account and we cannot verify your identity, we cannot
        grant access due to [GDPR](https://gdpr-info.eu/) privacy regulations.
        Image-Line Support Team
      - >-
        Title: Where is Toxic lll? (Legacy product)

        Answer: This article is about the legacy product Toxic 3, NOT about
        Toxic Biohazard. Please know that Toxic III has been discontinued in
        favor of Toxic Biohazard. Toxic III (read 3) is the predecessor of Toxic
        Biohazard.This outdated program is no longer available for use.
      - >-
        Title: What data does FL Studio collect and why?

        Answer: We want to be transparent about what data we collect (see "What
        data we collect with your permission and why" below). Rest assured, we
        do not collect or use your creative output - your music, scores, and
        sounds are entirely yours and remain securely on your machine. We are
        only interested in which parts of FL Studio and other tools you use to
        create your music plus what genre it is. This is so we can improve FL
        Studio to meet yours and other users needs. What data we collect with
        your permission and why See the [list

        here](https://docs.google.com/spreadsheets/d/1wSZY5Eah25ynIvKC67laUfr1vN3aeBidGAqKu-YzZMA/edit?usp=sharing)
        We understand your privacy concerns as the team at Image-Line also use
        software from other companies that ask for similar data from us. First
        and foremost we are a music production software company, not a spy
        agency! Our focus is on improving FL Studio - How are people using it
        and what features would help them. For transparency, you can review the
        list above to see what data we collect and why. We can't reach into FL
        Studio and see what you are doing in

        realtime, nor can we access anything else on your device. Nor do we want
        to. FL Studio sends the data listed above to our servers in a 'push'
        process at the moments listed in the spreadsheet. We get packets of text
        data, such as - Welcome window | Project starter Chord Progression
        option used | 12 April 2027. How would we use that? We analyze
        aggregated data from all users, not just you. For example, a query might
        be - For people who have used FL Studio for 3 months or less vs 3 months
        or more, how frequently do they use the Project starter Chord
        Progression tool. The result

        might be a graph that looks like this: Simulated Data From that we may
        conclude the Chord Tool is more interesting to new users, and so we
        might then focus more on making more features in it that are geared
        toward new users, or perhaps we would try to improve it in ways that
        keep its appeal after you have been using FL Studio for a while. Of
        course your data is voluntary and you can choose not to share it. Our
        legal obligations with respect to your data As a Belgian registered
        business, we are bound by European Union Law (EU). In particular this
        includes the [General Data

        Protection Regulation (GDPR).](https://gdpr.eu/) This regulation is
        designed to protect the privacy of EU citizens from unscrupulous
        behaviour by any organization that may want to collect information about
        you based in the EU. However, we extend our obligations under the GDPR
        here to all FL Studio customers, no matter where you are located! Please
        note, there are legitimate reasons allowed under the GDPR for us to
        collect identifiable data about you that are mutually agreed and
        beneficial. The GDPR acknowledges the following broad categories that
        apply to our case (although

        there are others): Consent - This is the most straightforward reason.
        Where your explicitly agrees for us to collect your data. The consent
        must be freely given, specific, informed, and unambiguous. Contract -
        The data collection is necessary to fulfill a contract with the
        customer, or to take steps before entering into a contract. For example,
        the FL Studio license includes a single user (you) and is for Lifetime
        Free Updates. This is a contract between us and we need to know who you
        are so that, among other reasons, if you lose access to your account
        (emails change) we can be sure we

        are dealing with the original contract participant. Vital Interests -
        The data is necessary to protect the vital interests of the customer or
        another person. This could be in situations preventing fraud. For
        example, some unscrupulous people use stolen Credit Cards to purchase FL
        Studio. By requiring validation of the ID of the user we can protect the
        original owners of those cards and ensure users lawfully purchased FL
        Studio. Legitimate Interests - This is the most flexible of the GDPR and
        applies to cases where we have a legitimate interest in processing your
        data. It is likely to be most

        appropriate where we use your data in ways you would reasonably expect
        and which have a minimal privacy impact, or where there is a compelling
        justification for the processing. Account recovery is one such example;
        We may ask you to identify yourself. While you will always have the
        right to decline, we will also have the right to refuse access to an
        account, where we can't be sure you are the legitimate custodian of the
        data in it (this is a privacy matter in itself).
  - source_sentence: >-
      What are the three main parameters of the Self Oscillating Filter (SOF) in
      the Drumpad instrument, and how does each parameter affect the sound
      produced?
    sentences:
      - >-
        Document_title: Edison 

        File_name: plugins/Edison_2.htm

        Heading_hierarchy: [Edison -> Exporting recorded audio to the Playlist]

        Anchor_id: [none]

        Alternatives for exporting audio from Edison to FL Studio are: Send
        sample to Playlist - Left-click - Sends the selection (or whole sample
        if no selection is made) to the Playlist as an [Audio
        Clip](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/../playlist_audioclip.htm)
        . 

        Right-click - Replaces the currently selected Audio Clip or Sampler
        Channel with the sample/selection in Edison. Drag / copy sample / move
        selection - Left-click on the button and drag to the desired location.
        The selected region in the Sample Edit
                Window (or whole sample if no selection is made) will be copied and moved to any compatible location in FL Studio, e.g. Sampler channels, Fruity Slicer, DirectWave, the Playlist, etc. Right-click to copy the selection to the clipboard. Tools > Sequencing • Send to Playlist as audio clip ( Shift
        +C )- Dumps the selection to the Playlist as an Audio Clip. • Send to
        selected Channel - Dumps the selection to the selected Channel. Save and
        load -
        [Save](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Edison_3.htm#Edison_3)
        the audio/selection to a file and

        re-import it through the
        [Browser](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/../browser.htm)
        . NOTES: Prior to exporting your audio to the Playlist use the Edit
        properties ( F2 )
        [dialog](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Ed

        ison_14.htm#Edison_14) so that
            Audio clips behave as expected: ➤ Clip auto-naming - Edison will use a default naming system, unless the audio clip is named in the sample [properties dialog](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Edison_14.htm#Edison_14) . ➤ Tempo sync - To set the
        Audio Clip so that it will be automatically stretched if the project
        tempo is changed (useful to keep rhythmic samples in time), click the
        'default' button in the 
              tempo settings panel. If the file is not set as 'tempo-sync' or has no tempo (0), it will not be stretched. ➤ 16 or 32-Bit? - If your soundcard was set to record at 16-Bit then it is reasonable to convert the file to 16-Bit prior to saving to disk, otherwise leave the file at 32-Bit float.
      - >-
        Document_title: Drumpad - Percussion Modeling Instrument

        File_name: plugins/Drumpad.htm

        Heading_hierarchy: [Drumpad - Percussion Modeling Instrument -> Self
        Oscillating Filter (SOF)]

        Anchor_id: [none]

        The SOF is an 'allpass' filter with a narrow frequency-peak exhibiting
        high levels of self-oscillation. When the frequency-peak is swept
        quickly downwards, it is useful to make kicks and snares. • Freq -
        Down-sweep starting frequency. The frequency will sweep down to ~20 Hz
        (fixed). • Decay -

        Down-sweep time. Fast sweeps produce the most realistic 'kick' sounds. •
        Level - SOF mix level. The SOF sweep is added before other filters in
        the modeling engine, immediately after the Membrane sound is generated.
      - >-
        Document_title: Dashboard

        File_name: plugins/Dashboard.htm

        Heading_hierarchy: [Dashboard]

        Anchor_id: [none]

        Dashboard is a legacy plugin , retained for compatibility with older
        projects. For new projects use [Control
        Surface](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/Control%20Surface.htm)
        . The Dashboard allows you to create software interfaces to control
        external

        MIDI hardware (keyboards, mixers, samplers, etc) or you can link
        controls to [internal
        targets](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/../automation_internal.htm)
        to bring together a range of commonly tweaked knobs in one place.
                  The Dashboard includes several pre-made panels (presets) for popular MIDI devices you can use directly in your projects. You can customize the existing panels for your own use or create new ones from scratch. In this way, you can adapt a Dashboard for any MIDI device you own
        without
                  using plugins dedicated to a specific hardware. Check out the range of customer created dashboards in the [Dashboard - Presets & Programming](https://support.image-line.com/redirect/dashboard_presets) forum.
  - source_sentence: >-
      Describe the role of the 'Decay' and 'Cutoff' controls in shaping the
      sound of the drum membrane in the Drumaxx instrument. How do these
      controls interact with the other parameters to influence the final sound
      output?
    sentences:
      - >-
        ➤ Noise - Special effect. Harmonics are sounded randomly. Use the RES to
        control the amount of randomness, width controls the length/duration of
        changes in harmonics. ➤ C

        ustom shape 1 & 2 - Define your own resonance filter shape with  'Filter
        (resonance) shape 1 & 2' envelopes. ➤ Low pass - Provides a resonance
        peak followed by a dip, this makes it an 'extra lowpass', so that the
        slope of the lowpass filter effectively becomes steeper as the resonance

        increases. • width - Resonance peak width, left = narrow, right = wide.
        ➤ adapt - Adaptive envelope mode. The bandwidth of the filter changes
        width as a function of the notes frequency, so that the resonance sounds
        similar regardless of the note pitch. Set by ear. • ofs - Offset shifts
        the

        resonant frequency +/- 2400 cents, relative to the cutoff frequency.
        NOTE: Resonance may become inaudible for low-pass modes if the offset
        moves the peak past the filter cutoff point, since
            there may be few or no frequencies to resonate beyond the cutoff point. • osc - Self-oscillation level. A sine wave is deliberately introduced at the resonant frequency developing more distinct 'tonal' sound. Can be useful when the offset moves the resonant peak beyond
            the filter cutoff point (see 'ofs' above).
      - >-
        Document_title: Wave Traveller

        File_name: plugins/Wave Traveller.htm

        Heading_hierarchy: [Wave Traveller -> Mute Regions in the Patch]

        Anchor_id: [none]

        You can select specific regions to be muted inside the current patch.
        Left-click and drag in the Mutes Definition Area ( 5 ). Black regions
        show muted areas. To unmute a region, Right-click and drag instead.
      - >-
        Document_title: Drumaxx - Percussion Modeling Instrument 

        File_name: plugins/Drumaxx.htm

        Heading_hierarchy: [Drumaxx - Percussion Modeling Instrument ->
        Membrane]

        Anchor_id: [none]

        The drum membrane (skin) has a significant impact on the tonal character
        of the final sound. There are 4 key parameters - Tension , Phase ,
        Material and Shape .
            When the right balance between these four controls is achieved, the sound will be most natural. We recommend adjusting Tension and Phase first, so that the pitch is close to
            what you need. Then adjust Material and Shape to fine-tune the tonal qualities. If the balance is right, the Drum Pad will sound like a drum, and if not, the Drum Pad can sound 'blurry' and 'loose'. • Decay - Decay time for the membrane response and reflections. • Tension - Membrane
        tension (how tightly stretched the drum is). • Material - This control
        moves the membrane material through a range of possible characteristics
        and types. Generally left are heavier slower materials and right are 
              lighter more reflective materials. Set by ear. • Cutoff - High frequency cutoff. • Phase - Relates to the impact-point on the drum-head. Set by ear. • Size - Drum-head size. • Shape - The shape can be varied from square (left) to round (right). Most drum-heads are round, but experimentation
        is always rewarded!
  - source_sentence: >-
      Describe the function of the "Absolute Pattern" option in the Riff
      Machine. How does enabling or disabling this option impact the slicing of
      notes in the Piano roll grid?
    sentences:
      - >-
        Document_title: Layer Settings

        File_name: chansettings_layer.htm

        Heading_hierarchy: [Layer Settings -> Options]

        Anchor_id: [none]

        • Levels Adjustment - This section contains controls for the volume
        (VOL) , panning (PAN) and Pitch of the linked layers. NOTE: The levels
        you set in the Layer Channel apply ONLY to the notes played through that
        layer. If you play a child of this Channel through 
                its own [Step Sequencer](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/channelrack.htm) dots or [Piano roll](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll.htm) , these settings will not be applied. • Layering section ➤ Set
        children - Assigns all selected Channels in the [Step
        Sequencer](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/channelrack.htm)
        as children in this Layer Channel. When you
                  play a note on the Layer Channel, all the children play along. To unassign a Channel from the Layer Channel, select all the Channels you want to remain 
                  as children and press the Set children button again (all unselected Channels become unassigned for this Layer Channel). ➤ Show children - Selects all Channels that are children of this Layer Channel in the Step Sequencer, and deselects all other Channels. ➤ Random - OFF: All children of
        the Layer Channel will sound on each note. ON: A single, random, Channel
        in the Layer will play. Use the 'Random' feature
                  to make more interesting percussion sounds, for example, by assigning many similar samples to each Channel in the Layer. This will give subtle variations on
                  each repeated note. ➤ Crossfade - ON: The Fade knob (below) will crossfade between two or more Channels in the Layer. ◆ Fade knob - Used to set the crossfade level in crossfade mode. For example; If you have 3 Layer Channels turning the Fade knob from left to right will 
                  sound: Child 1 > Child 1 + Child 2 > Child 2 > Child 2 + Child 3 > Child 3 . Channels are faded from top (knob left) to bottom (knob right) in the Channel Rack. NOTE: Crossfading only works with
                  FL Studio native format plugins, it does not work with VST/AU  plugins. • Sequential - ON: Each Channel will play in turn (round-robbin style) starting with the highest Channel working to the lowest when the ' Set children ' function was used. NOTE: The system remembers the Channel
        order when 'Set children' was used. To re-order the sequence, rearrange
        your Channels and reapply 'Set children'. • Layering menu - Click on the
        small arrow at the top left of this panel you can access some additional
        commands: ➤ Split children - Splits the children of the Layer Channel
        across

        the keyboard (starting with the root key of the Layer Channel),
        assigning each layer to a single key. The root keys of the children are
        automatically adjusted, so that the correct pitch is played through the
        Layer Channel. This feature is useful for creating drum kits or
        instruments where each

        note has different sample. ➤ Reset children - Resets the range and root
        notes for all Child Channels of a layer. Basically undoes the ' Split
        children ' actions. ➤ Group children - Adds all children of the Layer
        Channel to a group (a popup window will appear to enter the name of the
        group). For

        more information see the Channel Filtering section in the [Step
        Sequencer](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/channelrack.htm)
        page. ➤ Delete children - Removes selected children from the layer. •
        Preview Keyboard - The preview keyboard allows you

        to preview the Channel instrument (Left-clicking on the piano-keyboard),
        set the root key (Right-Click a key), and set key region (Left-click and
        drag on the ruler). See the [Miscellaneous Channel

        Settings](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/chansettings_misc.htm)
        page for more information on using the Preview Keyboard.
      - >-
        Document_title: Fruity Limiter + Compressor + Gate

        File_name: plugins/Fruity Limiter.htm

        Heading_hierarchy: [Fruity Limiter + Compressor + Gate -> Loudness]

        Anchor_id: [none]

        These controls affect the input and limit thresholds for Fruity Limiter.
        • GAIN (Purple) -  Limiter input gain. NOTE: this gain is
        post-Compressor and pre-Limiter. • SAT (Saturation) -  Controls the
        threshold level, signals reaching into the red region will be saturated.
        To lower the threshold, and so

        increase the saturation turn SAT left. Set by ear. Saturation (pleasing
        to some) is a type of [amplitude
        distortion](https://en.wikipedia.org/wiki/Distortion) , usually
        associated 
              with [Valve / Tube Amplifiers](https://en.wikipedia.org/wiki/Valve_sound) , tape or analog circuitry. Generally, as a waveform 
              exceeds the maximum amplitude that an analog system can carry, its shape is rounded or bent. This bending is a mild distortion that progressively increases as the input approaches the maximum (0 dB). TIP: For an alternative saturation sound allow the input to peak over the limit
        threshold, then blend in some '
        [Soft-clipping](https://en.wikipedia.org/wiki/Tube_sound#Soft_clipping)
        ' from the saturation. Over-limiting adds its own kind of saturation
        effect that interacts nicely with the formal saturation process. • CEIL
        (Ceiling) - Limiter ceiling. Sets the level above which

        the signal will be limited. While it is not possible to disable the
        limiter stage completely, setting the ceiling level to the maximum (+12
        dB)
              should effectively prevent limiter engaging. NOTE: For the Limiter to work the input signal must exceed the limiter threshold (CEIL). Your options are to lower the CEIL level, increase the GAIN level or increase the signal level entering the plugin.
      - >-
        Document_title: Riff Machine  

        File_name: pianoroll_riff_prog.htm

        Heading_hierarchy: [Riff Machine -> Options]

        Anchor_id: [none]

        These controls augment/modify the selected progression. Note that some
        controls will only have an effect if the original progression includes
        some variation in that parameter (PAN for example). • Time mul - Time
        multiplier, change the length of the notes. • PAN - Note panning
        multiplier. • VO

        L (Volume) - Note velocity multiplier. • MODX - Modulation X multiplier.
        • MODY - Modulation Y multiplier. • PITCH - Note pitch multiplier. •
        Absolute Pattern - On: any note slicing is based on the Piano roll grid.
        Off: Each note is sliced relative to its own start time. • Group notes -
        Groups any

        chopped-up notes, use the
        [grouping](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/toolbar_panels.htm#panel_shortcuticons_group)
        function on the [Toobar shortcut

        icons](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/toolbar_panels.htm#panel_shortcuticons)
        to activate note grouping. [Step 2. Chord
        Progression](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_chord.htm#Riff_Chord)
        [St

        ep 3.
        Arpeggiation](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_arp.htm#Riff_Arp)
        [Step 4. Mirroring
        Notes](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_mirror.htm#Riff_Mirror)
        [Step 5. Levels &

        Panning](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_levels.htm#Riff_Levels)
        [Step 6. Articulation (note
        length)](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_art.htm#Riff_Art)
        [Step 7. Groove (note

        timing)](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_groove.htm#Riff_Groove)
        [Step 8. Fit (note
        range)](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_fit.htm#Riff_Fit)
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
model-index:
  - name: SentenceTransformer based on Snowflake/snowflake-arctic-embed-l
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: Unknown
          type: unknown
        metrics:
          - type: cosine_accuracy@1
            value: 0.8788300835654597
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.9616991643454039
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.9818941504178273
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9895543175487466
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.8788300835654597
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.32056638811513466
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.19637883008356546
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09895543175487465
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.8788300835654597
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.9616991643454039
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.9818941504178273
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9895543175487466
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.939831343382967
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.9232421961356503
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.923862728471483
            name: Cosine Map@100

SentenceTransformer based on Snowflake/snowflake-arctic-embed-l

This is a sentence-transformers model finetuned from Snowflake/snowflake-arctic-embed-l. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: Snowflake/snowflake-arctic-embed-l
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 1024 dimensions
• Similarity Function: Cosine Similarity

Model Sources

• Documentation: Sentence Transformers Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("sentence_transformers_model_id")
# Run inference
sentences = [
    'Describe the function of the "Absolute Pattern" option in the Riff Machine. How does enabling or disabling this option impact the slicing of notes in the Piano roll grid?',
    'Document_title: Riff Machine  \nFile_name: pianoroll_riff_prog.htm\nHeading_hierarchy: [Riff Machine -> Options]\nAnchor_id: [none]\nThese controls augment/modify the selected progression. Note that some controls will only have an effect if the original progression includes some variation in that parameter (PAN for example). • Time mul - Time multiplier, change the length of the notes. • PAN - Note panning multiplier. • VO\nL (Volume) - Note velocity multiplier. • MODX - Modulation X multiplier. • MODY - Modulation Y multiplier. • PITCH - Note pitch multiplier. • Absolute Pattern - On: any note slicing is based on the Piano roll grid. Off: Each note is sliced relative to its own start time. • Group notes - Groups any\nchopped-up notes, use the [grouping](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/toolbar_panels.htm#panel_shortcuticons_group) function on the [Toobar shortcut\nicons](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/toolbar_panels.htm#panel_shortcuticons) to activate note grouping. [Step 2. Chord Progression](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_chord.htm#Riff_Chord) [St\nep 3. Arpeggiation](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_arp.htm#Riff_Arp) [Step 4. Mirroring Notes](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_mirror.htm#Riff_Mirror) [Step 5. Levels &\nPanning](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_levels.htm#Riff_Levels) [Step 6. Articulation (note length)](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_art.htm#Riff_Art) [Step 7. Groove (note\ntiming)](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_groove.htm#Riff_Groove) [Step 8. Fit (note range)](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll_riff_fit.htm#Riff_Fit)',
    "Document_title: Layer Settings\nFile_name: chansettings_layer.htm\nHeading_hierarchy: [Layer Settings -> Options]\nAnchor_id: [none]\n• Levels Adjustment - This section contains controls for the volume (VOL) , panning (PAN) and Pitch of the linked layers. NOTE: The levels you set in the Layer Channel apply ONLY to the notes played through that layer. If you play a child of this Channel through \n        its own [Step Sequencer](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/channelrack.htm) dots or [Piano roll](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/pianoroll.htm) , these settings will not be applied. • Layering section ➤ Set\nchildren - Assigns all selected Channels in the [Step Sequencer](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/channelrack.htm) as children in this Layer Channel. When you\n          play a note on the Layer Channel, all the children play along. To unassign a Channel from the Layer Channel, select all the Channels you want to remain \n          as children and press the Set children button again (all unselected Channels become unassigned for this Layer Channel). ➤ Show children - Selects all Channels that are children of this Layer Channel in the Step Sequencer, and deselects all other Channels. ➤ Random - OFF: All children of\nthe Layer Channel will sound on each note. ON: A single, random, Channel in the Layer will play. Use the 'Random' feature\n          to make more interesting percussion sounds, for example, by assigning many similar samples to each Channel in the Layer. This will give subtle variations on\n          each repeated note. ➤ Crossfade - ON: The Fade knob (below) will crossfade between two or more Channels in the Layer. ◆ Fade knob - Used to set the crossfade level in crossfade mode. For example; If you have 3 Layer Channels turning the Fade knob from left to right will \n          sound: Child 1 > Child 1 + Child 2 > Child 2 > Child 2 + Child 3 > Child 3 . Channels are faded from top (knob left) to bottom (knob right) in the Channel Rack. NOTE: Crossfading only works with\n          FL Studio native format plugins, it does not work with VST/AU  plugins. • Sequential - ON: Each Channel will play in turn (round-robbin style) starting with the highest Channel working to the lowest when the ' Set children ' function was used. NOTE: The system remembers the Channel\norder when 'Set children' was used. To re-order the sequence, rearrange your Channels and reapply 'Set children'. • Layering menu - Click on the small arrow at the top left of this panel you can access some additional commands: ➤ Split children - Splits the children of the Layer Channel across\nthe keyboard (starting with the root key of the Layer Channel), assigning each layer to a single key. The root keys of the children are automatically adjusted, so that the correct pitch is played through the Layer Channel. This feature is useful for creating drum kits or instruments where each\nnote has different sample. ➤ Reset children - Resets the range and root notes for all Child Channels of a layer. Basically undoes the ' Split children ' actions. ➤ Group children - Adds all children of the Layer Channel to a group (a popup window will appear to enter the name of the group). For\nmore information see the Channel Filtering section in the [Step Sequencer](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/channelrack.htm) page. ➤ Delete children - Removes selected children from the layer. • Preview Keyboard - The preview keyboard allows you\nto preview the Channel instrument (Left-clicking on the piano-keyboard), set the root key (Right-Click a key), and set key region (Left-click and drag on the ruler). See the [Miscellaneous Channel\nSettings](https://www.image-line.com/fl-studio-learning/fl-studio-online-manual/html/chansettings_misc.htm) page for more information on using the Preview Keyboard.",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Information Retrieval

• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.8788
cosine_accuracy@3	0.9617
cosine_accuracy@5	0.9819
cosine_accuracy@10	0.9896
cosine_precision@1	0.8788
cosine_precision@3	0.3206
cosine_precision@5	0.1964
cosine_precision@10	0.099
cosine_recall@1	0.8788
cosine_recall@3	0.9617
cosine_recall@5	0.9819
cosine_recall@10	0.9896
cosine_ndcg@10	0.9398
cosine_mrr@10	0.9232
cosine_map@100	0.9239

Training Details

Training Dataset

Unnamed Dataset

• Size: 5,776 training samples
• Columns: sentence_0 and sentence_1
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1
  type	string	string
  details	min: 12 tokens mean: 33.33 tokens max: 66 tokens	min: 37 tokens mean: 278.56 tokens max: 512 tokens

• Samples:

  sentence_0	sentence_1
  Explain the issue that arises with project names in FL Studio 20 when using non-English characters, and describe the steps needed to resolve this issue on a Windows 10 system.	Title: Projects names are not showing correctly. Names in non-english characters (Cyrillic Korean, Japanese, Chinese, Hindi, Thai, etc.). Answer: FL Studio 20 works in unicode and displays in Windows 10 automatically your local character set. However, for projects moved from older FL Studio program versions, FL Studio does not know the character set it needs to display. Language not set up correctly: Your FL Studio 20 program will look like this: Solution: correct language set up instructions: A. Please check this Windows support article: Follow the steps below to set up non-unicode language in windows 10 1. In search tab type "Region" and press enter. 2. In new window select "Administrative" 3. then click on "change system locale" 4. Select the language. B. Import your old projects again into FL Studio 20. The names will now show u...
  Discuss the importance of setting the correct language settings in FL Studio 20 for displaying project names accurately, especially when importing projects from older versions of the software.	Title: Projects names are not showing correctly. Names in non-english characters (Cyrillic Korean, Japanese, Chinese, Hindi, Thai, etc.). Answer: FL Studio 20 works in unicode and displays in Windows 10 automatically your local character set. However, for projects moved from older FL Studio program versions, FL Studio does not know the character set it needs to display. Language not set up correctly: Your FL Studio 20 program will look like this: Solution: correct language set up instructions: A. Please check this Windows support article: Follow the steps below to set up non-unicode language in windows 10 1. In search tab type "Region" and press enter. 2. In new window select "Administrative" 3. then click on "change system locale" 4. Select the language. B. Import your old projects again into FL Studio 20. The names will now show u...
  How can you toggle the visibility of the FL Studio window when using it as a ReWire client within Cubase SX™?	Document_title: Using FL Studio ReWire with Cubase SX™ File_name: rewire_client_cubase.htm Heading_hierarchy: [Using FL Studio ReWire with Cubase SX™ -> 5. Toggle the FL Studio window visibility] Anchor_id: [none] Clicking the FL Studio icon toggles the visibility of the FL Studio window inside Cubase™. If you need to hide the FL Studio window, use the close button in the FL Studio window (this will not terminate the current session) or click the icon button on the FL Studio ReWire panel. To display the window later, click the icon button again.

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 10
• per_device_eval_batch_size: 10
• num_train_epochs: 2
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 10
• per_device_eval_batch_size: 10
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 5e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1
• num_train_epochs: 2
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: False
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• eval_do_concat_batches: True
• fp16_backend: auto
• push_to_hub_model_id: None
• push_to_hub_organization: None
• mp_parameters:
• auto_find_batch_size: False
• full_determinism: False
• torchdynamo: None
• ray_scope: last
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• dispatch_batches: None
• split_batches: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin

Training Logs

Epoch	Step	Training Loss	cosine_ndcg@10
0.0865	50	-	0.9089
0.1730	100	-	0.9252
0.2595	150	-	0.9170
0.3460	200	-	0.9235
0.4325	250	-	0.9227
0.5190	300	-	0.9275
0.6055	350	-	0.9272
0.6920	400	-	0.9127
0.7785	450	-	0.9303
0.8651	500	0.0357	0.9303
0.9516	550	-	0.9353
1.0	578	-	0.9353
1.0381	600	-	0.9310
1.1246	650	-	0.9327
1.2111	700	-	0.9310
1.2976	750	-	0.9336
1.3841	800	-	0.9328
1.4706	850	-	0.9359
1.5571	900	-	0.9377
1.6436	950	-	0.9382
1.7301	1000	0.0074	0.9398
1.8166	1050	-	0.9398

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.3.1
• Transformers: 4.47.1
• PyTorch: 2.5.1+cu121
• Accelerate: 1.2.1
• Datasets: 3.2.0
• Tokenizers: 0.21.0

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}