Training in progress, step 100

README.md

@@ -16,7 +16,7 @@ model-index:
 <!-- 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/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/quirky_lats_at_mats/harmbench_adv_training/runs/4u15rhza)
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/adv-robustness/harmbench_adv_training/runs/hgvq89zv)
 # llama2-7b-sft-lora
 
 This model is a fine-tuned version of [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf) on the generator dataset.
@@ -43,10 +43,10 @@ The following hyperparameters were used during training:
 - eval_batch_size: 8
 - seed: 4
 - distributed_type: multi-GPU
-- num_devices: 4
+- num_devices: 8
 - gradient_accumulation_steps: 4
-- total_train_batch_size: 64
-- total_eval_batch_size: 32
+- total_train_batch_size: 128
+- total_eval_batch_size: 64
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: cosine
 - training_steps: 2000

adapter_config.json

@@ -16,12 +16,12 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "v_proj",
     "k_proj",
     "q_proj",
-    "up_proj",
     "o_proj",
-    "down_proj"
+    "v_proj",
+    "down_proj",
+    "up_proj"
   ],
   "task_type": "CAUSAL_LM"
 }

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4654ba892d076143ec5687141d3afb7b697a533f01f788e060d51bb4b2daf0d3
+oid sha256:42f7e5fd16674642b7f4bce7f8711f3751fc2bfa844082ec46f8fc1a8169d7b3
 size 258001832

runs/May20_23-02-38_9962c5574dd5/events.out.tfevents.1716246231.9962c5574dd5.35433.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3763a50a7d30b8132c214b544961e981bf2b1e7079c87fcaa0484440b564b26
-size 5051
+oid sha256:2536fba8c340bb09f954404653633a5d7b5fc5d0b357add481179694eb0438bd
+size 8285

step_0/README.md

@@ -203,4 +203,11 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
 ### Framework versions
 
 
+- PEFT 0.6.1
+## Training procedure
+
+
+### Framework versions
+
+
 - PEFT 0.6.1

step_0/adapter_config.json

@@ -16,12 +16,12 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "v_proj",
     "k_proj",
     "q_proj",
-    "up_proj",
     "o_proj",
-    "down_proj"
+    "v_proj",
+    "down_proj",
+    "up_proj"
   ],
   "task_type": "CAUSAL_LM"
 }

step_0/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d5a2bc4755dd43d7d40ba9e853eb1528b6f82c1de5eaf623707ae7753c17001
+oid sha256:9b752b40e6f4688c866de9e9675488d079afb968c4be67f29600b030b5b3fffa
 size 6200

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d5a2bc4755dd43d7d40ba9e853eb1528b6f82c1de5eaf623707ae7753c17001
+oid sha256:9b752b40e6f4688c866de9e9675488d079afb968c4be67f29600b030b5b3fffa
 size 6200