Push model using huggingface_hub.

README.md

@@ -26,7 +26,7 @@ You can then generate text as follows:
 ```python
 from transformers import pipeline
 
-generator = pipeline("text-generation", model="jvelja//tmp/tmpk54pbimq/jvelja/vllm-gemma2b-stringMatcher-newDataset_0")
+generator = pipeline("text-generation", model="jvelja//tmp/tmpsz1t8pzc/jvelja/vllm-gemma2b-stringMatcher-newDataset_0")
 outputs = generator("Hello, my llama is cute")
 ```
 
@@ -36,8 +36,8 @@ If you want to use the model for training or to obtain the outputs from the valu
 from transformers import AutoTokenizer
 from trl import AutoModelForCausalLMWithValueHead
 
-tokenizer = AutoTokenizer.from_pretrained("jvelja//tmp/tmpk54pbimq/jvelja/vllm-gemma2b-stringMatcher-newDataset_0")
-model = AutoModelForCausalLMWithValueHead.from_pretrained("jvelja//tmp/tmpk54pbimq/jvelja/vllm-gemma2b-stringMatcher-newDataset_0")
+tokenizer = AutoTokenizer.from_pretrained("jvelja//tmp/tmpsz1t8pzc/jvelja/vllm-gemma2b-stringMatcher-newDataset_0")
+model = AutoModelForCausalLMWithValueHead.from_pretrained("jvelja//tmp/tmpsz1t8pzc/jvelja/vllm-gemma2b-stringMatcher-newDataset_0")
 
 inputs = tokenizer("Hello, my llama is cute", return_tensors="pt")
 outputs = model(**inputs, labels=inputs["input_ids"])

adapter_config.json

@@ -20,8 +20,8 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "v_proj"
+    "v_proj",
+    "q_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14adfaf25441bb892e4d58c657e41770658dfe5a1a29cea501ef99e581ac8ebf
+oid sha256:091a63d95c47d4896f0dab289f033fadceb1dff21638fcb65fd782f3a5f442f5
 size 12793376

config.json

@@ -44,14 +44,14 @@
   "tracker_kwargs": {
     "wandb": {
       "name": "cv_gemma-2-2b-it_to_distilbert-base-uncased_EBS64_Joan",
-      "notes": "Dataset: cv\n    Same Prompt:  \n    Payload Prefixes: ['Movie Review: This movie was really amazing!', 'Movie Review: This movie was really terrible!']\n    Payload Template: Movie Review: This movie was really {payload}!\n    Separate Enc/Dec Data: True\n\n    Encoder: gemma-2-2b-it (LR: 2e-05)\n    Decoder: distilbert-base-uncased (LR: 3e-06)\n    Train Loop: v2_dylan\n\n    Effective Batch Sizes:\n    - Encoder: 64\n    - Decoder: 512\n\n    Training Iterations:\n    - Encoder updates: 100\n    - Decoder updates: 400\n    - Update Encoder First: False\n\n    Temperatures:\n    - Decoder Training: 1.0\n    - Encoder Training: 1.0\n    - Evaluation: 1.0\n\n    Encoder Parameters:\n    - KL Coefficient: 0.05\n    - LoRA: True\n    - Quantization: False\n    - Output Length: {'min': 42, 'max': 51}\n\n    Decoder Parameters:\n    - New Classification Head: True\n    - Use Probs Reward: False\n    - Weight Decay: 0.01\n    - Update Parameters: {'head': True, 'body': True}\n\n    Training Configuration:\n    - Update Encoder: True\n    - Update Decoder: True\n    - Paraphrase: False\n    - Leak Password: False\n    - WandB Logging: True\n    - Eval Every N: 50\n    - Number of Epochs: 100000\n\n    Debug:\n    - Override Dec Batch: False",
+      "notes": "Dataset: cv\n    Same Prompt:  \n    Payload Prefixes: ['Movie Review: This movie was really amazing!', 'Movie Review: This movie was really terrible!']\n    Payload Template: Movie Review: This movie was really {payload}!\n    Separate Enc/Dec Data: True\n\n    Encoder: gemma-2-2b-it (LR: 2e-05)\n    Decoder: distilbert-base-uncased (LR: 1e-05)\n    Train Loop: v2_dylan\n\n    Effective Batch Sizes:\n    - Encoder: 64\n    - Decoder: 512\n\n    Training Iterations:\n    - Encoder updates: 100\n    - Decoder updates: 400\n    - Update Encoder First: False\n\n    Temperatures:\n    - Decoder Training: 1.0\n    - Encoder Training: 1.0\n    - Evaluation: 1.0\n\n    Encoder Parameters:\n    - KL Coefficient: 0.05\n    - LoRA: True\n    - Quantization: False\n    - Output Length: {'min': 42, 'max': 51}\n\n    Decoder Parameters:\n    - New Classification Head: True\n    - Use Probs Reward: False\n    - Weight Decay: 0.01\n    - Update Parameters: {'head': True, 'body': True}\n\n    Training Configuration:\n    - Update Encoder: True\n    - Update Decoder: True\n    - Paraphrase: False\n    - Leak Password: False\n    - WandB Logging: True\n    - Eval Every N: 50\n    - Number of Epochs: 100000\n\n    Debug:\n    - Override Dec Batch: False",
       "tags": [
         "cv",
         "gemma-2-2b-it",
         "distilbert-base-uncased",
         "v2_dylan",
         "enc_lr_2e-05",
-        "dec_lr_3e-06",
+        "dec_lr_1e-05",
         "enc_eff_bs_64",
         "dec_eff_bs_512",
         "enc_updates_100",

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:216cdd88d033fb785cf1e59942fd17a97bb2f9ef37babb8deb819a0e8eb8dd26
+oid sha256:b683989d60948db17a83186ada9495a4ad7923e3c304beb5fba9388e3d638607
 size 10748