---
title: "Returns classifier models, thresholds, and training defaults"
method: GET
path: "/classifier/info"
tags: ["classifier"]
---

# Returns classifier models, thresholds, and training defaults

`GET /classifier/info`

## Response `200`

Successful Response

- ClassifierInfo
  - `weak_signal_threshold` number, required — Signal percentage below which training signal is weak
  - `strong_signal_threshold` number, required — Signal percentage above which training signal is strong
  - `transformer_models` ClassifierBaseModelOption[], required — Selectable transformer base models (span & document classification)
    - `value` string, required — HuggingFace model name
    - `label` string, required — Display label for the model
  - `embedding_models` ClassifierBaseModelOption[], required — Selectable embedding base models (sentence classification)
    - `value` string, required — HuggingFace model name
    - `label` string, required — Display label for the model
  - `training_params` ClassifierTrainingSettings, required
    - `lora_enabled` boolean, required — Whether to train with a LoRA adapter
    - `lora_rank` integer, required — Rank of the LoRA update matrices
    - `lora_alpha` integer, required — Scaling factor applied to LoRA updates
    - `lora_dropout` number, required — Dropout probability applied inside LoRA layers
    - `freeze_base_model` boolean, required — Freeze pretrained base-model weights. Without LoRA, only classifier layers are trained; LoRA requires this setting
    - `epochs` integer, required — Number of training epochs
    - `batch_size` integer, required — Training batch size
    - `early_stopping` boolean, required — Whether to use early stopping
    - `early_stopping_patience` integer, required — Number of validation epochs without improvement before stopping
    - `train_test_split` number, required — Fraction of selected training data reserved for validation
    - `base_learning_rate` number, required — Peak learning rate for the pretrained base model
    - `head_learning_rate` number, required — Peak learning rate for the classifier head and, when enabled, LoRA adapter parameters
    - `warmup_fraction` number, required — Fraction of optimizer steps used to increase each learning rate linearly from zero to its peak before linear decay
    - `weight_decay` number, required — Weight decay
    - `dropout` number, required — Model dropout rate
    - `chunk_size` integer, required — Token chunk size
    - `precision` union, required — Lightning training precision
      - 64 | 32 | 16
      - 'transformer-engine' | 'transformer-engine-float16' | '16-true' | '16-mixed' | 'bf16-true' | 'bf16-mixed' | '32-true' | '64-true'
      - '64' | '32' | '16' | 'bf16'
    - `averaging` 'micro' | 'macro', required

---

[API](https://skmtc.net/uhh-lt/apis/discourse-analysis-tool-suite-api.md) · [All operations](https://skmtc.net/uhh-lt/apis/discourse-analysis-tool-suite-api/llms.txt) · [OpenAPI document](https://skmtc-service-staging.skmtc.workers.dev/v1/apis/uhh-lt/discourse-analysis-tool-suite-api/revisions/6fcf6b0bc28a/schema)