ronit01/rag_tuned_minilm_mnr_100

SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. It maps sentences & paragraphs to a 384-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: sentence-transformers/all-MiniLM-L6-v2
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 384 dimensions
• Similarity Function: Cosine Similarity
• Supported Modality: Text

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'BertModel'})
  (1): Pooling({'embedding_dimension': 384, 'pooling_mode': 'mean', '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("ronit01/rag_tuned_minilm_mnr_100")
# Run inference
sentences = [
    'What must the create_model_fn return, and when is it invoked by RapidFire AI?',
    'Create Model Function\n------\n\nMandatory user-provided function to create HuggingFace model and tokenizer objects based on the  \nmodel type(s) and name(s) given in the :code:`RFModelConfig` and multi-config specification. \nAlso read :doc:`the LoRA and Model Configs page</models>`.\nA model can be imported from the Hugging Face model hub or read from a local checkpoint file. \n\nIt is passed to :func:`run_fit()` directly. Also read :doc:`the Experiment page</experiment>`.\n\nThis function is invoked when a trainer object is created for each run. \n\n\n.. py:function:: create_model_fn(model_config: Dict[str, Any]) -> Tuple[transformers.PreTrainedModel, transformers.PreTrainedTokenizer]\n\n  :param model_config: Dictionary injected by RapidFire AI into this user-defined function with all key-value pairs for one model config output by the config-group generator.\n  :type model_config: Dict[str, Any]\n\n  :return: Tuple containing the initialized Hugging Face model (e.g., ``AutoModelForCausalLM``, ``AutoModelForSequenceClassification``) and tokenizer (e.g., ``AutoTokenizer``, ``PreTrainedTokenizer``) objects\n  :rtype: Tuple[transformers.PreTrainedModel, transformers.PreTrainedTokenizer]\n\n\n**Example:**\n\n.. code-block:: python\n\n\t# From the SFT tutorial notebook\n\tdef sample_create_model(model_config):\n    \t"""Function to create model object for any given config; must return tuple of (model, tokenizer)"""\n\t\tfrom transformers import AutoModelForCausalLM, AutoTokenizer, AutoModelForSeq2SeqLM, AutoModelForMaskedLM\n\t\t\n\t\tmodel_name = model_config["model_name"]\n\t\tmodel_type = model_config["model_type"]\n\t\tmodel_kwargs = model_config["model_kwargs"]\n\t\t\n\t\tif model_type == "causal_lm":\n\t\t\tmodel = AutoModelForCausalLM.from_pretrained(model_name, **model_kwargs)\n\t\telif model_type == "seq2seq_lm":\n\t\t\tmodel = AutoModelForSeq2SeqLM.from_pretrained(model_name, **model_kwargs)\n\t\telif model_type == "masked_lm":\n\t\t\tmodel = AutoModelForMaskedLM.from_pretrained(model_name, **model_kwargs)\n\t\telif model_type == "custom":\n            # Handle custom model loading logic, e.g., loading your own checkpoints\n            # model = ... \n\t\t\tpass\n\t\telse:\n\t\t\t# Default to causal LM\n\t\t\tmodel = AutoModelForCausalLM.from_pretrained(model_name, **model_kwargs)\n\n\t\ttokenizer = AutoTokenizer.from_pretrained(model_name)\n\n\t\treturn (model,tokenizer)',
    'We currently support two common config group generators: :func:`RFGridSearch()` for grid search \nand :func:`RFRandomSearch()` for random search. \n\nMore support for AutoML heuristics such as SHA, HyperOpt, as well as an integration with \nthe popular AutoML library Optuna are coming soon. \nLikewise for RAG/context engineering, we also plan to support the AutoML heuristic syftr.\n\n\n.. py:function:: RFGridSearch(configs: Dict[str, Any] | List[Dict[str, Any]], trainer_type: str = "SFT" | "DPO" | "GRPO" | None)\n\n\t:param configs: A config dictionary with :func:`List()` for at least one knob; can be a list of such config dictionaries too.\n\t:type configs: Dict[str, Any] | List[Dict[str, Any]]\n\n\t:param trainer_type: The fine-tuning/post-training control flow to use: "SFT", "DPO", or "GRPO". Skip this argument for :func:`run_evals()`.\n\t:type trainer_type: str, optional \n\n\n.. py:function:: RFRandomSearch(configs: Dict[str, Any], trainer_type: str = "SFT" | "DPO" | "GRPO" | None, num_runs: int, seed: int = 42)\n\n\t:param configs: A config dictionary with :func:`List()` or :func:`Range()` for at least one knob.\n\t:type configs: Dict[str, Any]\n\n\t:param trainer_type: The fine-tuning/post-training control flow to use: "SFT", "DPO", or "GRPO". Skip this argument for :func:`run_evals()`.\n\t:type trainer_type: str, optional\n\n\t:param num_runs: Number of runs (full combinations of knob values) to sample in total.\n\t:type num_runs: int\n\n\t:param seed: Seed for random sampling of knob values to construct combinations. Default is 42.\n\t:type seed: int, optional \n\n\n**Notes:**\n\nFor :func:`RFGridSearch()`, each knob can have either a single value or a :func:`List()` of values but no knob \nshould have :func:`Range()` of values; otherwise, it will error out.\n\nFor :func:`RFRandomSearch()`, each knob can have either a single value, or a :func:`List()` of values, or a \n :func:`Range()` of values. The semantics of sampling are independently-identically-distributed (IID), i.e.,\nwe uniformly randomly pick a value from each discrete set and from each continuous set to construct the \nknob combination for one run. \nThen we repeat that sampling process in an IID way to accumulate :code:`num_runs` distinct combinations.\n\nNote that the return types of the config group generators are internal to RapidFire AI and they are usable only \nwithin the context of :func:`run_fit()` or :func:`run_evals()` in the :class:`Experiment` class.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000,  0.6946, -0.0594],
#         [ 0.6946,  1.0000, -0.0999],
#         [-0.0594, -0.0999,  1.0000]])

Training Details

Training Dataset

Unnamed Dataset

• Size: 52 training samples

• Columns: sentence_0 and sentence_1

• Approximate statistics based on the first 52 samples:

  	sentence_0	sentence_1
  type	string	string
  details	min: 11 tokens mean: 24.87 tokens max: 34 tokens	min: 31 tokens mean: 216.15 tokens max: 256 tokens

• Samples:

  sentence_0	sentence_1
  What are the four tabs on the main Experiments page of RapidFire AI's MLflow-fork dashboard?	The main "Experiments" page on the dashboard has 4 main tabs:

  • Table

  • Chart

  • Experiment Log

  • Interactive Control (IC) Log

  The screenshot below shows the "Table" view of an experiment with all its runs. Each run represents one model with one set of config knob values, which is standard dashboard semantics.

  .. raw:: html

  <img src="_static/mlflow-1-table.png" alt="Table view of runs metadata" 
       style="cursor: zoom-in; max-width: 100%;" onclick="this.requestFullscreen()">
  

  Metrics Plots

  The screenshot below shows the "Chart" view of an experiment with all its runs. Each plot corresponds to a metric, spanning :code:loss on the training set and evaluation set, as well all named metrics returned in your :func:compute_metrics() function in the trainer config.

  We call attention to 3 key aspects of the visualizations here:

  • The x-axis "Step" for the mini batch-level plots represents absolute number of minibatches seen by that run. So, if the :code:`bat... | | How do reward functions work in RapidFire AI for GRPO training, and what arguments does TRL inject into them? | Reward Functions

  ---

  User-provided reward function(s) needed for GRPO. You can create as many reward functions as you like with custom names.

  A list of such functions is passed to the :code:reward_funcs argument of :class:RFModelConfig. Also read: :doc:the LoRA and Model Configs page</models>. You can create multiple variants of this list with different subsets of functions and pass them all as a single :code:List to your :class:RFModelConfig to create a multi-config specification.

  These functions are invoked by the underlying HF trainer on the generated outputs on the fly.

  .. py:function:: reward_function(prompts, completions, completions_ids, trainer_state, **kwargs) -> List[float]

  :param prompts: List of input prompts that produced the completions.
  :type prompts: List[str] | List[List[Dict[str, str]]]
  
  :param completions: List of generated completions corresponding to above prompts.
  :type completions: List[str] | List[List[Dict[str, str]]]
  
  :param completi...</code> |
  

  | What are the five reward functions used in the GRPO tutorial for mathematical reasoning, and what does each one reward? | This tutorial shows Group Relative Policy Optimization (GRPO) to improve mathematical reasoning capabilities. GRPO is an RL approach that uses multiple reward functions to provide richer training signals.

  It uses the GSM8K mathematical reasoning dataset; see its details on Hugging Face <https://huggingface.co/datasets/openai/gsm8k>__. We use a sample of 500 training examples and 100 evaluation examples for tractable demo runtimes.

  The prompt format includes a system message instructing the model to respond with structured reasoning and answer tags, encouraging step-by-step mathematical problem solving with clear formatting.

  Model, Adapter, and Trainer Knobs

  We compare 3 different base model architectures: Llama-3.1-8B-Instruct, Qwen2.5-3B-Instruct, and Qwen2.5-7B-Instruct, all using 4-bit quantization for efficient training.

  All models use the same medium capacity LoRA configuration, targeting only 2 modules. We compare two different learning rates for the smaller Qw... |

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false,
      "directions": [
          "query_to_doc"
      ],
      "partition_mode": "joint",
      "hardness_mode": null,
      "hardness_strength": 0.0
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 100
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• do_predict: False
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• 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: 100
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_ratio: None
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• enable_jit_checkpoint: False
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• use_cpu: False
• seed: 42
• data_seed: None
• bf16: False
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: -1
• ddp_backend: None
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• group_by_length: False
• length_column_name: length
• project: huggingface
• trackio_space_id: trackio
• 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
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• hub_revision: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_for_metrics: []
• eval_do_concat_batches: True
• auto_find_batch_size: False
• full_determinism: False
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• include_num_input_tokens_seen: no
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: True
• use_cache: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

Training Time

• Training: 56.6 seconds

Framework Versions

• Python: 3.12.13
• Sentence Transformers: 5.4.1
• Transformers: 5.0.0
• PyTorch: 2.10.0+cu128
• Accelerate: 1.13.0
• Datasets: 4.0.0
• Tokenizers: 0.22.2

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{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}