Hugging Face's logo

HideIron
/
tiny-aya-global-sql-selfheal

PEFT
Safetensors
text-to-sql
lora
self-healing
agenttune
Model card Files
xet
 Community
HideIron commited on
Commit
f4e7866
·
verified ·
1 Parent(s): e58b933

DPO LoRA (153 self-heal pairs)

Browse files
Files changed (6) hide show
  1. adapter_config.json +3 -3
  2. adapter_model.safetensors +1 -1
  3. dpo_config.yaml +2 -2
  4. lexsi_provenance.json +1 -1
  5. training_args.bin +1 -1
  6. training_stats.json +11 -11
adapter_config.json CHANGED
@@ -32,13 +32,13 @@
32
  "rank_pattern": {},
33
  "revision": null,
34
  "target_modules": [
35
- "gate_proj",
36
  "up_proj",
37
  "v_proj",
38
  "down_proj",
39
- "k_proj",
40
  "q_proj",
41
- "o_proj"
 
 
42
  ],
43
  "target_parameters": null,
44
  "task_type": "CAUSAL_LM",
 
32
  "rank_pattern": {},
33
  "revision": null,
34
  "target_modules": [
 
35
  "up_proj",
36
  "v_proj",
37
  "down_proj",
 
38
  "q_proj",
39
+ "o_proj",
40
+ "k_proj",
41
+ "gate_proj"
42
  ],
43
  "target_parameters": null,
44
  "task_type": "CAUSAL_LM",
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3eb100f6a1ba40281fa146fa4152718bd1f7660c00a20fe543fd8e3eb59e8db8
3
  size 120981200
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a528834f81fda89e10c18f016b3d2f1f76b5e9eef7009a5df79f10c6e3ef542a
3
  size 120981200
dpo_config.yaml CHANGED
@@ -72,7 +72,7 @@ label_names: null
72
  label_smoothing: 0.0
73
  label_smoothing_factor: 0.0
74
  ld_alpha: null
75
- learning_rate: 3.0e-05
76
  length_column_name: length
77
  liger_kernel_config: null
78
  load_best_model_at_end: false
@@ -97,7 +97,7 @@ model_init_kwargs:
97
  dtype: bfloat16
98
  trust_remote_code: false
99
  neftune_noise_alpha: null
100
- num_train_epochs: 4.0
101
  optim: adamw_torch_fused
102
  optim_args: null
103
  optim_target_modules: null
 
72
  label_smoothing: 0.0
73
  label_smoothing_factor: 0.0
74
  ld_alpha: null
75
+ learning_rate: 2.0e-05
76
  length_column_name: length
77
  liger_kernel_config: null
78
  load_best_model_at_end: false
 
97
  dtype: bfloat16
98
  trust_remote_code: false
99
  neftune_noise_alpha: null
100
+ num_train_epochs: 5.0
101
  optim: adamw_torch_fused
102
  optim_args: null
103
  optim_target_modules: null
lexsi_provenance.json CHANGED
@@ -3,7 +3,7 @@
3
  "library": "agenttune",
4
  "version": "1.1.5",
5
  "git_sha": null,
6
- "created_at": "2026-11-13T23:29:43Z",
7
  "base_model": "CohereLabs/tiny-aya-global",
8
  "method": "agentic.dpo",
9
  "inputs": [],
 
3
  "library": "agenttune",
4
  "version": "1.1.5",
5
  "git_sha": null,
6
+ "created_at": "2026-11-14T00:16:37Z",
7
  "base_model": "CohereLabs/tiny-aya-global",
8
  "method": "agentic.dpo",
9
  "inputs": [],
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:71a7e4ef6b0f929696decc4f7293dd6b9f664aa8bc020eef34b59f731aa21981
3
  size 5969
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c8841aa6a77ac8e31283ab3670933fecceaf7cf123791689de23fdbb83590fec
3
  size 5969
training_stats.json CHANGED
@@ -5,22 +5,22 @@
5
  "use_rollouts": false,
6
  "agentic_mode": false,
7
  "tools": [],
8
- "train_size": 158,
9
  "eval_size": 0,
10
- "final_loss": 0.2593421244993806,
11
- "total_steps": 40,
12
  "training_history": [],
13
  "metrics": {
14
- "train_runtime": 1269.7545,
15
- "train_samples_per_second": 0.498,
16
  "train_steps_per_second": 0.032,
17
- "total_flos": 2.141871945331507e+16,
18
- "train_loss": 0.2593421244993806,
19
- "epoch": 4.0
20
  },
21
  "config_kwargs": {
22
  "model": "CohereLabs/tiny-aya-global",
23
- "train_dataset": "Dataset({\n features: ['prompt', 'chosen', 'rejected'],\n num_rows: 158\n})",
24
  "output_dir": "/tmp/out/adapter",
25
  "peft_config": {
26
  "r": 16,
@@ -37,10 +37,10 @@
37
  "down_proj"
38
  ]
39
  },
40
- "num_train_epochs": 4.0,
41
  "per_device_train_batch_size": 1,
42
  "gradient_accumulation_steps": 16,
43
- "learning_rate": 3e-05,
44
  "beta": 0.1,
45
  "seed": 42,
46
  "bf16": true,
 
5
  "use_rollouts": false,
6
  "agentic_mode": false,
7
  "tools": [],
8
+ "train_size": 153,
9
  "eval_size": 0,
10
+ "final_loss": 0.3636344361305237,
11
+ "total_steps": 50,
12
  "training_history": [],
13
  "metrics": {
14
+ "train_runtime": 1543.1982,
15
+ "train_samples_per_second": 0.496,
16
  "train_steps_per_second": 0.032,
17
+ "total_flos": 2.622984019611648e+16,
18
+ "train_loss": 0.3636344361305237,
19
+ "epoch": 5.0
20
  },
21
  "config_kwargs": {
22
  "model": "CohereLabs/tiny-aya-global",
23
+ "train_dataset": "Dataset({\n features: ['prompt', 'chosen', 'rejected'],\n num_rows: 153\n})",
24
  "output_dir": "/tmp/out/adapter",
25
  "peft_config": {
26
  "r": 16,
 
37
  "down_proj"
38
  ]
39
  },
40
+ "num_train_epochs": 5.0,
41
  "per_device_train_batch_size": 1,
42
  "gradient_accumulation_steps": 16,
43
+ "learning_rate": 2e-05,
44
  "beta": 0.1,
45
  "seed": 42,
46
  "bf16": true,