Paramir commited on
Commit
63128a6
·
verified ·
1 Parent(s): 59d120a

Fix model card base model metadata

Browse files
Files changed (1) hide show
  1. README.md +12 -7
README.md CHANGED
@@ -1,6 +1,5 @@
1
  ---
2
  license: other
3
- library_name: transformers
4
  pipeline_tag: text-classification
5
  tags:
6
  - promptguard
@@ -9,8 +8,10 @@ tags:
9
  - dlp
10
  - korean
11
  - roberta
 
 
 
12
  base_model:
13
- - Qwen/Qwen3-Embedding-0.6B
14
  - klue/roberta-base
15
  ---
16
 
@@ -20,12 +21,14 @@ This repository hosts the runtime model artifacts for PromptGuard's context-risk
20
 
21
  ## Runtime Architecture
22
 
23
- PromptGuard uses a two-stage context classifier:
24
 
25
- 1. Qwen embedding + One-vs-Rest Logistic Regression candidate generator
26
  2. Fine-tuned KLUE RoBERTa label-aware verifier
27
 
28
- The Logistic Regression stage is a high-recall candidate generator. The RoBERTa verifier is the final accept/reject model for candidate labels.
 
 
29
 
30
  ## Files
31
 
@@ -80,13 +83,15 @@ PROMPTGUARD_VERIFIER_RUNTIME_ENABLED=true
80
  PROMPTGUARD_VERIFIER_MANIFEST_PATH=/opt/promptguard/models/context_lr_roberta_active_best_f1_manifest.json
81
  ```
82
 
 
 
83
  ## Data and Privacy
84
 
85
- This repository contains runtime artifacts only. It does not include raw prompts, source documents, extracted text, filenames, or training/evaluation detail rows.
86
 
87
  ## License Notes
88
 
89
- The PromptGuard application code is distributed separately on GitHub. These model artifacts are published for PromptGuard runtime use. `Qwen/Qwen3-Embedding-0.6B` declares Apache-2.0 metadata on Hugging Face. `klue/roberta-base` does not expose a license field in its Hugging Face model card metadata at the time this card was written, so downstream users should review base-model and training-data obligations before redistribution or commercial use.
90
 
91
  ## Limitations
92
 
 
1
  ---
2
  license: other
 
3
  pipeline_tag: text-classification
4
  tags:
5
  - promptguard
 
8
  - dlp
9
  - korean
10
  - roberta
11
+ - sklearn
12
+ - logistic-regression
13
+ - joblib
14
  base_model:
 
15
  - klue/roberta-base
16
  ---
17
 
 
21
 
22
  ## Runtime Architecture
23
 
24
+ PromptGuard uses a two-stage context classifier. This repository is a runtime artifact package, not a single Transformer checkpoint loaded from the repository root.
25
 
26
+ 1. Frozen Qwen embedding dependency + One-vs-Rest Logistic Regression candidate generator
27
  2. Fine-tuned KLUE RoBERTa label-aware verifier
28
 
29
+ The Logistic Regression stage is a high-recall linear candidate generator stored as a `joblib` artifact. It is not a fine-tuned Qwen model and does not have its own Hugging Face base model. It expects normalized embedding vectors produced by `Qwen/Qwen3-Embedding-0.6B`.
30
+
31
+ The RoBERTa verifier is the fine-tuned Transformer stage. Its base model is `klue/roberta-base`, and the fine-tuned verifier checkpoint is stored under `models/context_verifier_klue_roberta_base_lrmined_v287_global002_compactv2_lpft_focal_1p2ep/`.
32
 
33
  ## Files
34
 
 
83
  PROMPTGUARD_VERIFIER_MANIFEST_PATH=/opt/promptguard/models/context_lr_roberta_active_best_f1_manifest.json
84
  ```
85
 
86
+ Do not load this repository root directly with `AutoModelForSequenceClassification.from_pretrained("OASecure/promptguard-context-classifier")`. PromptGuard reads `models/context_lr_roberta_active_best_f1_manifest.json`, loads the LR `joblib` file for candidate generation, and loads the verifier directory named in that manifest for RoBERTa verification.
87
+
88
  ## Data and Privacy
89
 
90
+ This repository contains runtime artifacts only. It does not include raw prompts, source documents, extracted text, filenames, training/evaluation detail rows, embedding vectors, logits, exact classifier scores, or full masked prompts.
91
 
92
  ## License Notes
93
 
94
+ The PromptGuard application code is distributed separately on GitHub. These model artifacts are published for PromptGuard runtime use. `Qwen/Qwen3-Embedding-0.6B` is used as a frozen embedding dependency for the LR stage, not as the fine-tuned base model of this repository. `klue/roberta-base` is the verifier base model. Downstream users should review base-model and training-data obligations before redistribution or commercial use.
95
 
96
  ## Limitations
97