Config file

The configuration file controls the training pipeline through five main sections: base, tuning, evaluation, synthgen, and trace_processing. Each section handles a specific aspect of the model training process.

File format

The configuration file supports two formats depending on how you interact with distil labs:

• Webapp: Use JSON format (.json file)
• API: Use YAML format (.yaml file)

Both formats are functionally equivalent—choose based on your workflow. Examples in this documentation show YAML, but the JSON equivalent is straightforward:

# YAML (API)
base:
  task: question-answering
  student_model_name: Llama-3.2-3B-Instruct

// JSON (Webapp)
{
  "base": {
    "task": "question-answering",
    "student_model_name": "Llama-3.2-3B-Instruct"
  }
}

Configuration structure

base:
  # General parameters (task is required)
  task: classification

tuning:
  # Fine-tuning parameters
  num_train_epochs: 32

evaluation:
  # Evaluation parameters
  num_few_shot_examples: 1

synthgen:
  # Synthetic data generation parameters
  generation_target: 10000

trace_processing:
  # Trace processing parameters
  relabel: true

Base configuration

General parameters relevant to the overall task.

Parameter	Type	Default	Description
task	string	required	Type of NLP task to be solved. See supported task types below.
student_model_name	string	Llama-3.2-1B-Instruct	Base model to use for the student model. This is the model we finetune for your use-case.
teacher_model_name	string	openai.gpt-oss-120b	Teacher model used to generate synthetic data and from which we distil knowledge.
random_seed	integer | null	123	Random seed used across distillib for reproducible random sampling.
llm_num_parallel_requests	integer	4	Maximum number of LLM requests to send in parallel across the teacher, synthgen, and judge pipelines. Set to 1 to disable parallelism.

Supported task types

Task	Value	Description
Question Answering	question-answering	Extract or generate answers from text based on queries
Classification	classification	Assign text to categories from a fixed set
Tool Calling	tool-calling-closed-book	Select and invoke functions based on user requests
Multi-turn Tool Calling	multi-turn-tool-calling-closed-book	Handle multi-step conversations with function calls
Open Book QA (RAG)	question-answering-open-book	Answer questions using provided context passages
Closed Book QA	question-answering-closed-book	Answer questions using knowledge learned during training

Supported student models

Model	Value
Llama 3.2 1B Instruct	Llama-3.2-1B-Instruct
Llama 3.2 3B Instruct	Llama-3.2-3B-Instruct
Llama 3.1 8B Instruct	Llama-3.1-8B-Instruct
SmolLM2 135M	SmolLM2-135M-Instruct
SmolLM2 1.7B	SmolLM2-1.7B-Instruct
FunctionGemma 270M	functiongemma-270m-it
Gemma 3 270M	gemma-3-270m-it
Gemma 3 1B	gemma-3-1b-it
Gemma 3 4B	gemma-3-4b-it
Qwen3 0.6B	Qwen3-0.6B
Qwen3 1.7B	Qwen3-1.7B
Qwen3 4B	Qwen3-4B-Instruct-2507
Qwen3 8B	Qwen3-8B
Liquid LFM2 350M	LFM2-350M
Liquid LFM2 1.2B	LFM2-1.2B
Liquid LFM2 2.6B	LFM2-2.6B
Liquid LFM2.5 350M	LFM2.5-350M
Liquid LFM2.5 1.2B Instruct	LFM2.5-1.2B-Instruct

Supported teacher models

Model	Value
GPT OSS 120B	openai.gpt-oss-120b
GPT OSS 120B Thinking	openai.gpt-oss-120b-thinking
GPT OSS 20B	openai.gpt-oss-20b
GPT OSS 20B Thinking	openai.gpt-oss-20b-thinking
DeepSeek V3.1	deepseek.v3.1
Qwen3 235B A22B	Qwen3-235B-A22B-Instruct-2507
Qwen3 480B A35B Coder	Qwen3-480B-A35B-Coder
Qwen2.5 VL 72B	Qwen2.5-VL-72B-Instruct
ZAI GLM 5	zai.glm-5
ZAI GLM 5 Thinking	zai.glm-5-thinking
Moonshot Kimi K2 Thinking	moonshotai.kimi-k2-thinking
Moonshot Kimi K2.5	moonshotai.kimi-k2.5
MiniMax M2 Thinking	minimax.minimax-m2-thinking

---

Tuning configuration

Parameters controlling the finetuning of the student model.

Parameter	Type	Default	Description
learning_rate	float	5e-5	The initial learning rate for AdamW optimizer.
learning_rate_scheduler	string	linear	The scheduler type to use. Options: cosine, linear, constant.
weight_decay	float	0.0	Weight decay applied to all layers except bias and LayerNorm weights in AdamW optimizer.
warmup_ratio	float	0.05	Ratio of total training steps used for linear warmup from 0 to learning_rate.
bf16	boolean	true	Whether to use bf16 16-bit (mixed) precision training instead of 32-bit training.
use_lora	boolean	true	Whether to use LoRA for student training.
lora_r	integer	64	LoRA attention dimension (rank). Only used if use_lora is true.
lora_alpha_multiplier	integer	1	Alpha parameter for LoRA scaling is lora_r * lora_alpha_multiplier. Only used if use_lora is true.
per_device_train_batch_size	integer	1	Batch size per GPU/device for training.
per_device_eval_batch_size	integer	1	Batch size per GPU/device for evaluation.
num_train_epochs	integer	4	Total number of training epochs.
train_eval_split	float	0.2	Fraction of training data used for evaluation. Must be between 0 and 1 (exclusive).
gradient_accumulation_steps	integer	1	Number of update steps to accumulate gradients before performing a backward/update pass. Effectively multiplies the batch size by this factor without increasing memory usage.
num_few_shot_examples_student	integer	0	Number of few-shot examples when running student evaluation and tuning. If above 0, at least one example per class is used for classification tasks.
memory_optimized_training	boolean	false	Enable activation offloading and gradient checkpointing to reduce GPU memory usage at the cost of significantly slower training. Only enable this if training runs out of GPU memory.
use_qlora	boolean	false	Load the base model in 4-bit NF4 (QLoRA) during finetuning, then attach LoRA adapters in higher precision. Reduces base-model VRAM by roughly 3x at the cost of slightly slower training. Only takes effect when use_lora is true. Requires bitsandbytes (Linux only).

RLVR (Reinforcement Learning with Verifiable Rewards)

RLVR is an optional reinforcement learning stage that runs after SFT finetuning. It uses reward signals from an LLM-as-a-judge to further improve model performance. Set rlvr_dataset_size to a value greater than 0 to enable it.

Parameter	Type	Default	Description
rlvr_dataset_size	float	0.0	Proportion of the dataset to use for the RLVR split. Must be between 0.0 and 1.0. Default 0.0 means RLVR is disabled.
rlvr_llm_as_a_judge_model_name	string	openai.gpt-oss-120b	Model used to power the LLM-as-a-judge for RLVR reward signals.
rlvr_per_device_batch_size	integer	6	Batch size per GPU/device for RLVR training and evaluation. Must be a multiple of rlvr_num_generations.
rlvr_num_generations	integer	6	Number of generations per prompt during RLVR training.
rlvr_num_train_epochs	integer	1	Number of training epochs for RLVR finetuning.

---

Evaluation configuration

Parameters used in teacher evaluation.

Parameter	Type	Default	Description
num_few_shot_examples	integer	1	Number of few-shot examples when running teacher evaluation. If above 0, at least one example per class is used for classification tasks.
llm_as_a_judge_model_name	string	openai.gpt-oss-120b	Model used to power the LLM-as-a-judge evaluation.
expand_tool_calling_turns	boolean	true	If true, each line in multi-turn tool calling test files is expanded into multiple evaluation lines, each ending at a tool call.

---

Synthetic generation configuration

Parameters for fine-grained control over synthetic data generation.

Parameter	Type	Default	Description
generation_target	integer	10000	Target number of synthetic examples to generate. For Closed-Book QA, this is calculated as len(unstructured_data) * generation_per_unstructured_context.
generation_in_single_call	integer	4	Number of examples to generate per teacher/LLM invocation.
generation_iteration_size	integer	128	Batch size for the generate-validate cycle.
generation_per_unstructured_context	integer | null	null	Examples to generate per unstructured context. Only used with question-answering-closed-book task. Overwrites generation_target when set.
num_positive_exemplars_per_generation	integer	2	Number of in-context examples for the class/task being generated.
num_negative_exemplars_per_generation	integer	2	Number of in-context examples for classes not being generated. Only used for classification tasks.
num_unlabelled_exemplars_per_generation	integer	2	Number of unlabelled examples provided during each teacher invocation.
validation_max_total_length	integer	10000	Maximum total length (input + output) of generated examples in characters.
validation_similarity_threshold	float	0.95	Similarity threshold for deduplication. Generated data with similarity above this threshold to seed data are removed.
teacher_temperature	float	0.7	Temperature for teacher output. Controls balance between predictability and creativity. Must be between 0.0 and 1.0.
teacher_max_tokens	integer	32000	Maximum number of tokens in the generated response. Kept well below typical model context limits so the reserved output budget does not crowd out large (e.g. multi-image) prompts.
match_generated_distribution_to_seed	boolean	false	Match generated data class distribution to seed data. Only used for classification tasks.
num_distractor_context_blocks	integer	0	Number of distractor context blocks per example. Setting above zero enables RAFT training.
output_is_json	boolean	false	Only generate synthetic data with valid JSON outputs. Only relevant for QA tasks.
basic_mutators_to_use	list[string]	["complexity"]	List of basic mutators to use for data generation. Supported options: complexity, length, specificity.
mutation_topics	list[list[string]] | list[string]	[]	Selection of topics to sample from to guide the generation process.

---

Trace processing configuration

Parameters for the trace processing pipeline, which converts production traces into training and testing data.

Parameter	Type	Default	Description
relabel	boolean	true	If true, use a committee of models to relabel trace examples. If false, use the original labels from traces.
relevance_filtering	boolean	true	If true, score each trace with an LLM and drop those below the relevance / coherence thresholds. If false, relevance filtering is skipped entirely and every seed trace flows straight to the next step.
relevance_filtering_batch_size	integer	32	Number of examples scored per batch during relevance filtering.
min_relevance_score	integer	4	Minimum relevance score (1-5) for a trace to pass relevance filtering.
min_coherence_score	integer	3	Minimum coherence score (1-5) for a trace to pass coherence filtering. Lower values allow more corrupted traces through for committee repair.
num_traces_as_training_base	integer	200	Number of traces to use as the seed for generating training examples. Unused traces beyond this count are used as unstructured data.
num_traces_as_testing_base	integer	200	Number of traces to use as the seed for generating testing examples. Unused traces beyond this count are used as unstructured data. Ignored if a test set is provided.
min_generated_examples	integer	20	Minimum number of examples that trace processing must produce. Raises an error if fewer are generated, to prevent training with too few examples.
max_unstructured	integer	10000	Maximum number of unstructured data examples to include.
observation_format	string	openai_messages	Format of trace observations in traces.jsonl. Options: langfuse (Langfuse observation objects with id, input, output), openai_messages (objects with a messages array of chat completion messages), openai_messages_with_images (OpenAI messages that may include images), unstructured_with_openai_messages (unstructured data with OpenAI messages).
remove_system_prompt_from_traces	boolean	true	If true, strip leading system messages from traces (before unstructured export) and from processed examples. Defaults to true because the system prompt is typically captured by the job description, and keeping it in the conversation breaks the single-turn [user, assistant] shape expected at the training boundary.
compress_job_description	boolean	false	If true, compress the job description using the teacher model before relevance filtering. Useful when the task description is very long and would overwhelm the filtering LLM.
teacher_model_name	string	zai.glm-5	Teacher model used for relevance filtering and picking the best relabelled answer from the committee.
relabelling_committee_models	list[string]	[]	If the list is non-empty, models in the list are used to produce candidate relabels. Each model generates an output for every example and the trace processing teacher aggregates them into the final relabel. Only used when relabel is true.

---

Example configuration

Minimal configuration

base:
  task: question-answering
  student_model_name: Llama-3.2-3B-Instruct
  teacher_model_name: openai.gpt-oss-120b

Full configuration example

base:
  task: question-answering-open-book
  student_model_name: Qwen3-1.7B
  teacher_model_name: openai.gpt-oss-120b
  random_seed: 42

tuning:
  learning_rate: 1e-4
  learning_rate_scheduler: cosine
  use_lora: true
  lora_r: 32
  num_train_epochs: 3
  train_eval_split: 0.15

evaluation:
  num_few_shot_examples: 2

synthgen:
  generation_target: 5000
  generation_in_single_call: 8
  teacher_temperature: 0.6
  validation_similarity_threshold: 0.9

trace_processing:
  relabel: true
  num_traces_as_training_base: 5000
  num_traces_as_testing_base: 100

---

Model-specific notes

Reasoning teacher models

Reasoning teacher models (the GPT OSS, DeepSeek, GLM, Kimi, and MiniMax families) require a teacher temperature between 0.5 and 0.7. Configurations that set synthgen.teacher_temperature outside this range will raise a validation error.

GPT OSS 120B Thinking

The openai.gpt-oss-120b-thinking model uses a medium reasoning effort setting by default for enhanced chain-of-thought capabilities.

Tool Calling

Tool calling tasks have specific model compatibility requirements:

Student models: Only Qwen3, Llama 3-family, LFM2/LFM2.5, and FunctionGemma models are supported for tool-calling-closed-book and multi-turn-tool-calling-closed-book tasks.

Teacher models for multi-turn: Multi-turn tool calling (multi-turn-tool-calling-closed-book) requires a teacher from the tested tool-calling allowlist. Common choices include:

• openai.gpt-oss-120b / openai.gpt-oss-120b-thinking
• openai.gpt-oss-20b
• Qwen3-235B-A22B-Instruct-2507
• zai.glm-5 / zai.glm-5-thinking
• moonshotai.kimi-k2-thinking / moonshotai.kimi-k2.5
• minimax.minimax-m2-thinking