Upload folder using huggingface_hub

README.md

@@ -0,0 +1,15 @@
+---
+license: mit
+tags:
+- any-to-any
+- omega
+- omegalabs
+- bittensor
+- agi
+---
+
+This is an Any-to-Any model checkpoint for the OMEGA Labs x Bittensor Any-to-Any subnet.
+
+Check out the [git repo](https://github.com/omegalabsinc/omegalabs-anytoany-bittensor) and find OMEGA on X: [@omegalabsai](https://x.com/omegalabsai).
+
+Trained by pruning

adapter_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49804b152a1353ebec41a685f4eac17b38e09e006f2d51d4cedd110668a54563
+size 13673786

meta_model_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7acf319d3b563a1e73e8ccfc6e8dbfcee3244224752cf5af7137c9889244674f
+size 16219158403

training_config.yml

@@ -0,0 +1,75 @@
+model:
+  _component_: models.lora_mmllama3_8b
+  lora_attn_modules:
+  - q_proj
+  - v_proj
+  apply_lora_to_mlp: false
+  apply_lora_to_output: false
+  lora_rank: 16
+  lora_alpha: 32
+  perception_tokens: 2
+  use_clip: false
+tokenizer:
+  _component_: models.a2a_tokenizer
+  path: models/tokenizer.model
+checkpointer:
+  _component_: torchtune.utils.FullModelMetaCheckpointer
+  checkpoint_dir: checkpoints/Meta-Llama-3-8B-Instruct/original
+  checkpoint_files:
+  - consolidated.00.pth
+  adapter_checkpoint: null
+  recipe_checkpoint: null
+  output_dir: output_checkpoints/omega_a2a
+  model_type: LLAMA3
+resume_from_checkpoint: false
+interim_checkpoint_steps: 5000
+interim_gen_steps: null
+max_new_tokens: 100
+temperature: 0.6
+top_k: 225
+dataset:
+  _component_: ds.EvenBatcher
+  buffer_size: 64
+  dataset:
+    _component_: ds.RoundRobinDataset
+    datasets:
+    - _component_: ds.CaptionInstructDataset
+      dataset_path: ds/sam_llava/output.parquet
+      train_on_input: false
+seed: null
+shuffle: true
+batch_size: 1
+optimizer:
+  _component_: torch.optim.AdamW
+  weight_decay: 0.001
+  lr: 1.0e-05
+lr_scheduler:
+  _component_: torchtune.modules.get_cosine_schedule_with_warmup
+  num_warmup_steps: 100
+loss:
+  _component_: torch.nn.CrossEntropyLoss
+epochs: 4
+max_steps_per_epoch: null
+gradient_accumulation_steps: 32
+compile: false
+output_dir: /tmp/lora_finetune_output
+metric_logger:
+  _component_: torchtune.utils.metric_logging.StdoutLogger
+  log_dir: /dev/stdout
+log_every_n_steps: 8
+device: cuda
+dtype: bf16
+enable_activation_checkpointing: false
+profiler:
+  _component_: torchtune.utils.profiler
+  enabled: false
+inference:
+  prompt_template: 'Video:
+
+    {video}
+
+    Caption the previous video.'
+  max_new_tokens: 300
+  temperature: 0.6
+  top_k: 300
+  quantizer: null