eryk-mazus
/

polka-1.1b

@@ -1,230 +1,30 @@
 ---
 base_model: eryk-mazus/tinyllama-with-custom-tokenizer
-tags:
-- generated_from_trainer
-model-index:
-- name: workspace/tmp/
-  results: []
----
-<!-- This model card has been generated automatically according to the information the Trainer had access to. You
-should probably proofread and complete it, then remove this comment. -->
-[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
-<details><summary>See axolotl config</summary>
-axolotl version: `0.3.0`
-```yaml
-base_model: eryk-mazus/tinyllama-with-custom-tokenizer
-model_type: LlamaForCausalLM
-tokenizer_type: AutoTokenizer
-is_llama_derived_model: true
-load_in_8bit: false
-load_in_4bit: false
-strict: false
 datasets:
-  - path: eryk-mazus/polka-pretrain-en-pl-v1
-    type: completion # format from earlier
-    field: text # Optional[str] default: text, field to use for completion data
-dataset_prepared_path:
-val_set_size: 0.05
-output_dir: /workspace/tmp/
-sequence_len: 2048
-sample_packing: false
-adapter:
-lora_model_dir:
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_fan_in_fan_out:
-wandb_project: polka
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-gradient_accumulation_steps: 2
-micro_batch_size: 4
-num_epochs: 1
-lr_scheduler:
-learning_rate: 0.00005
-optimizer: adamw_torch
-adam_beta1: 0.9
-adam_beta2: 0.95
-adam_epsilon: 0.00001
-max_grad_norm: 1.0
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-warmup_steps: 0
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-eval_steps: 1000
-save_steps: 1000
-save_total_limit: 2
-debug:
-deepspeed:
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"
 ```
-</details><br>
-# workspace/tmp/
-This model is a fine-tuned version of [eryk-mazus/tinyllama-with-custom-tokenizer](https://huggingface.co/eryk-mazus/tinyllama-with-custom-tokenizer) on the None dataset.
-It achieves the following results on the evaluation set:
-- Loss: 1.8795
-## Model description
-More information needed
-## Intended uses & limitations
-More information needed
-## Training and evaluation data
-More information needed
-## Training procedure
-### Training hyperparameters
-The following hyperparameters were used during training:
-- learning_rate: 5e-05
-- train_batch_size: 4
-- eval_batch_size: 4
-- seed: 42
-- distributed_type: multi-GPU
-- num_devices: 8
-- gradient_accumulation_steps: 2
-- total_train_batch_size: 64
-- total_eval_batch_size: 32
-- optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
-- lr_scheduler_type: cosine
-- num_epochs: 1
-### Training results
-| Training Loss | Epoch | Step  | Validation Loss |
-|:-------------:|:-----:|:-----:|:---------------:|
-| 3.0469        | 0.01  | 1000  | 3.0497          |
-| 2.664         | 0.02  | 2000  | 2.6586          |
-| 2.5018        | 0.04  | 3000  | 2.4944          |
-| 2.5955        | 0.05  | 4000  | 2.3988          |
-| 2.2783        | 0.06  | 5000  | 2.3338          |
-| 2.3171        | 0.07  | 6000  | 2.2852          |
-| 2.189         | 0.08  | 7000  | 2.2459          |
-| 2.3594        | 0.09  | 8000  | 2.2153          |
-| 2.1882        | 0.11  | 9000  | 2.1882          |
-| 2.2699        | 0.12  | 10000 | 2.1659          |
-| 2.1273        | 0.13  | 11000 | 2.1469          |
-| 2.1041        | 0.14  | 12000 | 2.1291          |
-| 2.1698        | 0.15  | 13000 | 2.1138          |
-| 2.2126        | 0.16  | 14000 | 2.1004          |
-| 2.1065        | 0.18  | 15000 | 2.0886          |
-| 2.0589        | 0.19  | 16000 | 2.0764          |
-| 2.0537        | 0.2   | 17000 | 2.0663          |
-| 1.9746        | 0.21  | 18000 | 2.0569          |
-| 2.2128        | 0.22  | 19000 | 2.0477          |
-| 2.1342        | 0.23  | 20000 | 2.0393          |
-| 2.0643        | 0.25  | 21000 | 2.0312          |
-| 2.2776        | 0.26  | 22000 | 2.0240          |
-| 1.94          | 0.27  | 23000 | 2.0173          |
-| 1.8249        | 0.28  | 24000 | 2.0111          |
-| 1.966         | 0.29  | 25000 | 2.0049          |
-| 1.9351        | 0.31  | 26000 | 1.9994          |
-| 1.9563        | 0.32  | 27000 | 1.9947          |
-| 1.9496        | 0.33  | 28000 | 1.9878          |
-| 2.0127        | 0.34  | 29000 | 1.9835          |
-| 2.0043        | 0.35  | 30000 | 1.9794          |
-| 2.0227        | 0.36  | 31000 | 1.9748          |
-| 1.9308        | 0.38  | 32000 | 1.9704          |
-| 1.9183        | 0.39  | 33000 | 1.9655          |
-| 1.9919        | 0.4   | 34000 | 1.9620          |
-| 1.9351        | 0.41  | 35000 | 1.9580          |
-| 1.9103        | 0.42  | 36000 | 1.9537          |
-| 1.7521        | 0.43  | 37000 | 1.9512          |
-| 1.9567        | 0.45  | 38000 | 1.9454          |
-| 2.022         | 0.46  | 39000 | 1.9426          |
-| 1.8526        | 0.47  | 40000 | 1.9398          |
-| 1.8912        | 0.48  | 41000 | 1.9370          |
-| 2.0546        | 0.49  | 42000 | 1.9334          |
-| 2.0607        | 0.5   | 43000 | 1.9308          |
-| 2.0078        | 0.52  | 44000 | 1.9279          |
-| 1.889         | 0.53  | 45000 | 1.9253          |
-| 1.8587        | 0.54  | 46000 | 1.9222          |
-| 1.8571        | 0.55  | 47000 | 1.9199          |
-| 1.8806        | 0.56  | 48000 | 1.9178          |
-| 1.8483        | 0.58  | 49000 | 1.9150          |
-| 1.7862        | 0.59  | 50000 | 1.9130          |
-| 1.8989        | 0.6   | 51000 | 1.9102          |
-| 1.9389        | 0.61  | 52000 | 1.9083          |
-| 1.9301        | 0.62  | 53000 | 1.9065          |
-| 1.9522        | 0.63  | 54000 | 1.9046          |
-| 1.883         | 0.65  | 55000 | 1.9027          |
-| 1.9647        | 0.66  | 56000 | 1.9002          |
-| 1.9284        | 0.67  | 57000 | 1.8988          |
-| 1.8836        | 0.68  | 58000 | 1.8974          |
-| 1.8472        | 0.69  | 59000 | 1.8956          |
-| 2.1232        | 0.7   | 60000 | 1.8945          |
-| 1.8571        | 0.72  | 61000 | 1.8933          |
-| 1.8043        | 0.73  | 62000 | 1.8918          |
-| 1.9468        | 0.74  | 63000 | 1.8906          |
-| 1.9173        | 0.75  | 64000 | 1.8896          |
-| 1.7762        | 0.76  | 65000 | 1.8880          |
-| 2.032         | 0.77  | 66000 | 1.8876          |
-| 1.9362        | 0.79  | 67000 | 1.8867          |
-| 1.8308        | 0.8   | 68000 | 1.8854          |
-| 1.9289        | 0.81  | 69000 | 1.8847          |
-| 1.9467        | 0.82  | 70000 | 1.8841          |
-| 1.8798        | 0.83  | 71000 | 1.8835          |
-| 1.8868        | 0.84  | 72000 | 1.8828          |
-| 1.8905        | 0.86  | 73000 | 1.8820          |
-| 1.9508        | 0.87  | 74000 | 1.8816          |
-| 1.7983        | 0.88  | 75000 | 1.8813          |
-| 1.7693        | 0.89  | 76000 | 1.8806          |
-| 1.7371        | 0.9   | 77000 | 1.8804          |
-| 1.8705        | 0.92  | 78000 | 1.8802          |
-| 1.8707        | 0.93  | 79000 | 1.8799          |
-| 1.9113        | 0.94  | 80000 | 1.8799          |
-| 2.1314        | 0.95  | 81000 | 1.8797          |
-| 1.9132        | 0.96  | 82000 | 1.8795          |
-| 2.0349        | 0.97  | 83000 | 1.8796          |
-| 1.7939        | 0.99  | 84000 | 1.8795          |
-| 1.8357        | 1.0   | 85000 | 1.8795          |
-### Framework versions
-- Transformers 4.36.2
-- Pytorch 2.1.2+cu121
-- Datasets 2.16.1
-- Tokenizers 0.15.0

 ---
+license: apache-2.0
 base_model: eryk-mazus/tinyllama-with-custom-tokenizer
 datasets:
+- allenai/MADLAD-400
+- eryk-mazus/polka-pretrain-en-pl-v1
+language:
+- pl
+- en
+pipeline_tag: text-generation
+---
+![image/png](https://cdn-uploads.huggingface.co/production/uploads/61bf0e11c88f3fd22f654059/EMSrPEzAFkjY9nvbaJoC3.png)
+# Polka-1.1b
+`polka-1.1b` takes the [TinyLlama-1.1B](https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T) model and enhances it by continuing pretraining on an additional **5.7 billion Polish tokens**, primarily sourced from the [MADLAD-400](https://arxiv.org/abs/2309.04662) dataset. The tokens were sampled in a 10:1 ratio between Polish and English shards using [DSIR](https://github.com/p-lambda/dsir). Furthermore, Polka extends the TinyLlama tokenizer's vocabulary to 43,882 tokens, improving its efficiency for generating Polish text.
+The training took 425 RTX 4090 GPU hours on a single 8 x RTX 4090 machine with DeepSpeed ZeRO-2.
+## Notes
+...
+## Sample code
+```python
+...
 ```