DPO LoRA (158 self-heal pairs)

adapter_config.json

@@ -32,13 +32,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "o_proj",
-    "v_proj",
+    "gate_proj",
     "up_proj",
-    "q_proj",
+    "v_proj",
     "down_proj",
-    "gate_proj",
-    "k_proj"
+    "k_proj",
+    "q_proj",
+    "o_proj"
   ],
   "target_parameters": null,
   "task_type": "CAUSAL_LM",

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e51a2d5343ab17b2aec9a27df6064efc3f424f1ab1b459c5e502dac9c6fd539
+oid sha256:3eb100f6a1ba40281fa146fa4152718bd1f7660c00a20fe543fd8e3eb59e8db8
 size 120981200

dpo_config.yaml

@@ -72,7 +72,7 @@ label_names: null
 label_smoothing: 0.0
 label_smoothing_factor: 0.0
 ld_alpha: null
-learning_rate: 1.0e-05
+learning_rate: 3.0e-05
 length_column_name: length
 liger_kernel_config: null
 load_best_model_at_end: false
@@ -90,14 +90,14 @@ loss_weights: null
 lr_scheduler_kwargs: null
 lr_scheduler_type: linear
 max_grad_norm: 1.0
-max_length: 1536
+max_length: 2176
 max_steps: -1
 metric_for_best_model: null
 model_init_kwargs:
   dtype: bfloat16
   trust_remote_code: false
 neftune_noise_alpha: null
-num_train_epochs: 3.0
+num_train_epochs: 4.0
 optim: adamw_torch_fused
 optim_args: null
 optim_target_modules: null

lexsi_provenance.json

@@ -3,7 +3,7 @@
   "library": "agenttune",
   "version": "1.1.5",
   "git_sha": null,
-  "created_at": "2026-11-13T22:51:02Z",
+  "created_at": "2026-11-13T23:29:43Z",
   "base_model": "CohereLabs/tiny-aya-global",
   "method": "agentic.dpo",
   "inputs": [],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d13dcbb4aa28ec125885000e68c64e23a6965141510dd3574807cae812d5920
+oid sha256:71a7e4ef6b0f929696decc4f7293dd6b9f664aa8bc020eef34b59f731aa21981
 size 5969

training_stats.json

@@ -7,16 +7,16 @@
   "tools": [],
   "train_size": 158,
   "eval_size": 0,
-  "final_loss": 0.556226130326589,
-  "total_steps": 30,
+  "final_loss": 0.2593421244993806,
+  "total_steps": 40,
   "training_history": [],
   "metrics": {
-    "train_runtime": 880.787,
-    "train_samples_per_second": 0.538,
-    "train_steps_per_second": 0.034,
-    "total_flos": 1.5379329064329216e+16,
-    "train_loss": 0.556226130326589,
-    "epoch": 3.0
+    "train_runtime": 1269.7545,
+    "train_samples_per_second": 0.498,
+    "train_steps_per_second": 0.032,
+    "total_flos": 2.141871945331507e+16,
+    "train_loss": 0.2593421244993806,
+    "epoch": 4.0
   },
   "config_kwargs": {
     "model": "CohereLabs/tiny-aya-global",
@@ -37,10 +37,10 @@
         "down_proj"
       ]
     },
-    "num_train_epochs": 3.0,
+    "num_train_epochs": 4.0,
     "per_device_train_batch_size": 1,
     "gradient_accumulation_steps": 16,
-    "learning_rate": 1e-05,
+    "learning_rate": 3e-05,
     "beta": 0.1,
     "seed": 42,
     "bf16": true,
@@ -52,7 +52,7 @@
       "dtype": "bfloat16",
       "trust_remote_code": false
     },
-    "max_prompt_length": 1408,
-    "max_length": 1536
+    "max_prompt_length": 2048,
+    "max_length": 2176
   }
 }