Training in progress, step 10000

README.md

@@ -1,7 +1,7 @@
 ---
 base_model: gpt2
 datasets:
-- distily/filtered_redpajama_en
+- distily/c4_multilingual_1M
 library_name: Distily
 license: creativeml-openrail-m
 tags:
@@ -18,7 +18,7 @@ model-index:
 
 Distilled with [Distily](https://github.com/lapp0/distily) library
 using teacher model [gpt2](https://huggingface.co/gpt2)
-on dataset [distily/filtered_redpajama_en](https://huggingface.co/datasets/distily/filtered_redpajama_en).
+on dataset [distily/c4_multilingual_1M](https://huggingface.co/datasets/distily/c4_multilingual_1M).
 
 <!-- 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.
@@ -78,13 +78,13 @@ GPT2LMHeadModel(
 
 # Resource Usage
 
-- Max Train VRAM Use: 15.7006 GB
-- Available VRAM: 23.6429 GB
+- Max Train VRAM Use: 15.7133 GB
+- Available VRAM: 23.4329 GB
 - GPUs: 
   - 1x NVIDIA GeForce RTX 4090
-- CPUs: 32
-- CPU Memory: 61.9353 GB
-- CPU Memory Bandwidth: 800 GB/s
+- CPUs: 64
+- CPU Memory: 251.7299 GB
+- CPU Memory Bandwidth: 1600 GB/s
 
 # Distillation (Teacher -> Student) Architecture Difference:
 
@@ -115,7 +115,7 @@ GPT2LMHeadModel(
 <br/>
 
 # Train Dataset
-Trained on 652,439,618 tokens from the [distily/filtered_redpajama_en](https://huggingface.co/datasets/distily/filtered_redpajama_en) dataset.
+Trained on 448,532,986 tokens from the [distily/c4_multilingual_1M](https://huggingface.co/datasets/distily/c4_multilingual_1M) dataset.
 
 - Num Samples: `998,000`
 - Subset: `None`
@@ -154,7 +154,7 @@ The following hyperparameters were used during training:
 - eval_batch_size: `8`
 - seed: `42`
 - optimizer: `Adam with betas=(0.9,0.999) and epsilon=1e-08`
-- lr_scheduler_type: `polynomial`
+- lr_scheduler_type: `constant`
 - num_epochs: `1.0`
 - distillation_objective: `DistillationObjective(
     logits_loss_component=LossComponent(
@@ -172,7 +172,7 @@ The following hyperparameters were used during training:
         projector='orthogonal'
     )
 )`
-- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7fb3590067a0>`
+- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7a7b4c69df60>`
 - student_model_name_or_path: `None`
 - student_config_name_or_path: `distilbert/distilgpt2`
 - student_model_config: `None`
@@ -182,12 +182,14 @@ The following hyperparameters were used during training:
 - teacher_model_name_or_path: `gpt2`
 - teacher_load_in_8bit: `False`
 - teacher_load_in_4bit: `False`
-- dataset_uri: `distily/filtered_redpajama_en`
+- dataset_uri: `distily/c4_multilingual_1M`
 - dataset_subset: `None`
 - dataset_split: `train`
 - dataset_column_name: `text`
 - dataset_sample_size: `1000000`
 - dataset_test_size: `0.002`
+- dataset_shuffle: `True`
+- dataset_shuffle_seed: `42`
 - gradient_accumulation_steps: `1`
 - weight_decay: `0.0`
 - max_grad_norm: `1.0`

logs/dataset_shuffle=True, dataset_split=train, dataset_subset=None, dataset_uri=distily_synth_gpt2_t1_seq_1M, lr_scheduler_kwargs=None, lr_scheduler_type=constant/events.out.tfevents.1725760207.b57c76173204

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3e525f097a6934a9b308761f599b1ae4850e00ce40b5e400fb414e488874fea1
-size 162422
+oid sha256:82c4980a456e19c3e6b470f4cc180d082e6fa3624717f76e4b1743def4624093
+size 319882

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70a6a98d8412de8e5f0df7a9ab442e0def3d38c81ee2e6d449523a271fc1312a
+oid sha256:0850e401c88cd6a58139f1e4526c3d931c9e18728f54b414dec45a0f15fc413f
 size 163832792

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dfaedc86c93e6f13587735b9be3aa15fa7027d42fab93c396b2259e3cbb33add
-size 5496
+oid sha256:6acba741f2c88cefc2e9c1f928d277d97e465c091be0f52ed282a068e0b6a6bd
+size 5560