chrisjcc commited on
Commit
d70d592
·
verified ·
1 Parent(s): 1e738b2

Upload folder using huggingface_hub

Browse files
README.md CHANGED
@@ -13,7 +13,7 @@ tags:
13
  - ambiguity-detection
14
  - question-answering
15
  - qwen2.5
16
- - sft
17
  - reinforcement-learning
18
  ---
19
 
@@ -31,12 +31,12 @@ the missing information and ask a targeted clarification question.
31
  This repository contains the exact W&B artifact approved by the project's
32
  model promotion procedure.
33
 
34
- - **Training method:** Supervised Fine-Tuning (SFT)
35
- - **Model variant:** `sft`
36
  - **W&B Registry alias at promotion:** `production`
37
- - **W&B artifact:** `wandb-registry-Model/AskBeforeAnswer-Models:v0`
38
- - **W&B artifact digest:** `8dd12fd8a79a7f5be7eac5d94def44c1`
39
- - **Release:** `v0.1.5`
40
  - **Base model:** `Qwen/Qwen2.5-7B-Instruct`
41
  - **Dataset:** `chrisjcc/ask-before-answer-dataset`
42
 
@@ -80,7 +80,27 @@ the promotion record.
80
 
81
  Evaluation
82
 
83
- Evaluation results are maintained in the project evaluation artifacts.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
84
 
85
  ## Usage
86
  ```
 
13
  - ambiguity-detection
14
  - question-answering
15
  - qwen2.5
16
+ - dpo
17
  - reinforcement-learning
18
  ---
19
 
 
31
  This repository contains the exact W&B artifact approved by the project's
32
  model promotion procedure.
33
 
34
+ - **Training method:** Direct Preference Optimization (DPO)
35
+ - **Model variant:** `dpo`
36
  - **W&B Registry alias at promotion:** `production`
37
+ - **W&B artifact:** `wandb-registry-Model/AskBeforeAnswer-dpo:v0`
38
+ - **W&B artifact digest:** `51fadb59065e11213173e689bd606b0f`
39
+ - **Release:** `v0.2.0`
40
  - **Base model:** `Qwen/Qwen2.5-7B-Instruct`
41
  - **Dataset:** `chrisjcc/ask-before-answer-dataset`
42
 
 
80
 
81
  Evaluation
82
 
83
+ ## LLM-as-a-Judge Evaluation Leaderboard
84
+
85
+ The following scores were computed using W&B Weave with a
86
+ Gemini-based judge scorer on a randomly selected
87
+ **50-sample** subset
88
+ of the `sewon/ambig_qa` (validation split).
89
+
90
+ | Metric | base | dpo_only | sft | sft_dpo | clarifier_lora | orpo | grpo |
91
+ |:----------------------|----------:|-----------:|---------:|----------:|-----------------:|---------:|---------:|
92
+ | ambiguity_detection | 0.966 | 0.974 | 0.97 | 0.946 | 0.994 | 0.968 | 0.952 |
93
+ | clarification_quality | 0.784 | 0.79 | 0.794 | 0.778 | 0.796 | 0.796 | 0.792 |
94
+ | usefulness | 0.88 | 0.882 | 0.896 | 0.878 | 0.896 | 0.896 | 0.894 |
95
+ | model_accuracy | 0.62 | 0.62 | 0.6 | 0.62 | 0.6 | 0.64 | 0.62 |
96
+ | clarify_precision | 0.617021 | 0.617021 | 0.631579 | 0.657143 | 0.6 | 0.636364 | 0.634146 |
97
+ | clarify_recall | 0.966667 | 0.966667 | 0.8 | 0.766667 | 1 | 0.933333 | 0.866667 |
98
+ | clarify_f1 | 0.753247 | 0.753247 | 0.705882 | 0.707692 | 0.75 | 0.756757 | 0.732394 |
99
+ | action_f1_answer | 0.173913 | 0.173913 | 0.375 | 0.457143 | 0 | 0.307692 | 0.344828 |
100
+ | macro_f1 | 0.46358 | 0.46358 | 0.540441 | 0.582418 | 0.375 | 0.532225 | 0.538611 |
101
+ | answer_accuracy | 0.05 | 0.05 | 0 | 0 | 0 | 0.1 | 0.1 |
102
+ | facet_generation_rate | 0.0851064 | 0.234043 | 1 | 1 | 0 | 0.977273 | 1 |
103
+ | clarify_ratio | 1.56667 | 1.56667 | 1.26667 | 1.16667 | 1.66667 | 1.46667 | 1.36667 |
104
 
105
  ## Usage
106
  ```
adapter_config.json CHANGED
@@ -35,13 +35,13 @@
35
  "rank_pattern": {},
36
  "revision": null,
37
  "target_modules": [
38
- "q_proj",
39
- "down_proj",
40
  "gate_proj",
41
- "k_proj",
42
  "o_proj",
 
43
  "v_proj",
44
- "up_proj"
 
45
  ],
46
  "target_parameters": null,
47
  "task_type": "CAUSAL_LM",
 
35
  "rank_pattern": {},
36
  "revision": null,
37
  "target_modules": [
 
 
38
  "gate_proj",
39
+ "up_proj",
40
  "o_proj",
41
+ "down_proj",
42
  "v_proj",
43
+ "q_proj",
44
+ "k_proj"
45
  ],
46
  "target_parameters": null,
47
  "task_type": "CAUSAL_LM",
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:44531fd80e249b7cbf4fc124522fc7f6be0b0202382eb865bb0e2c1590b469c1
3
  size 161533192
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6ff988941d4722799874703b10d7938908ad0304e61b8da8b0cc265d1bfece2b
3
  size 161533192
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1d5e8814a1cbbbf820e4191c98a514605ae36e48454ddd9eb02b6e69b79b2049
3
- size 6008
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c131f5192230a49a82dd678ab22fe1f78e9d22d4f73185693c6c49559b73fe52
3
+ size 6584