File

First experiment worked so great i opted for release, even though material was unclean. Remarakably! stable on transformers

README.md

@@ -1,3 +1,25 @@
 ---
-license: wtfpl
+library_name: peft
 ---
+## Training procedure
+
+### Framework versions
+
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+- PEFT 0.5.0
+
+- PEFT 0.5.0

adapter_config.json

@@ -0,0 +1,20 @@
+{
+  "auto_mapping": null,
+  "base_model_name_or_path": "models/gpt2",
+  "bias": "none",
+  "fan_in_fan_out": true,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "lora_alpha": 24,
+  "lora_dropout": 0.05,
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 16,
+  "revision": null,
+  "target_modules": [
+    "c_attn"
+  ],
+  "task_type": "CAUSAL_LM"
+}

adapter_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2bcdf24fe7b797ec3670bc6864cc51ce92fde2fa9e35a81b8475ee6350c1fa70
+size 2367673

training_log.json

@@ -0,0 +1,16 @@
+{
+  "base_model_name": "gpt2",
+  "base_model_class": "GPT2LMHeadModel",
+  "base_loaded_in_4bit": false,
+  "base_loaded_in_8bit": false,
+  "projections": "c_attn",
+  "loss": 0.8766,
+  "learning_rate": 0.00013416407864998736,
+  "epoch": 15.0,
+  "current_steps": 677,
+  "train_runtime": 33.5196,
+  "train_samples_per_second": 5.37,
+  "train_steps_per_second": 0.447,
+  "total_flos": 41438875484160.0,
+  "train_loss": 0.8892534891764323
+}

training_parameters.json

@@ -0,0 +1,30 @@
+{
+  "lora_name": "mbtk.py",
+  "always_override": true,
+  "save_steps": 0.0,
+  "micro_batch_size": 2,
+  "batch_size": 96,
+  "epochs": 48.0,
+  "learning_rate": "3e-4",
+  "lr_scheduler_type": "inverse_sqrt",
+  "lora_rank": 16,
+  "lora_alpha": 24,
+  "lora_dropout": 0.05,
+  "cutoff_len": 448,
+  "dataset": "None",
+  "eval_dataset": "None",
+  "format": "None",
+  "eval_steps": 100.0,
+  "raw_text_file": "einmbtkreaktorkannnichteinfach",
+  "overlap_len": 96,
+  "newline_favor_len": 0,
+  "higher_rank_limit": false,
+  "warmup_steps": 100.0,
+  "optimizer": "adamw_torch",
+  "hard_cut_string": "/0",
+  "train_only_after": "",
+  "stop_at_loss": 0,
+  "add_eos_token": false,
+  "min_chars": 0.0,
+  "report_to": "None"
+}

training_prompt.json

@@ -0,0 +1,3 @@
+{
+  "template_type": "raw_text"
+}