初始化项目，由ModelHub XC社区提供模型
Model: esmaeil-abedi-dev/verifier-ce-qwen2.5-0.5b Source: Original Platform
This commit is contained in:
ModelHub XC
36
.gitattributes
vendored
Normal file
36
.gitattributes
vendored
Normal file
@@ -0,0 +1,36 @@
|
||||
*.7z filter=lfs diff=lfs merge=lfs -text
|
||||
*.arrow filter=lfs diff=lfs merge=lfs -text
|
||||
*.bin filter=lfs diff=lfs merge=lfs -text
|
||||
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
||||
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
||||
*.ftz filter=lfs diff=lfs merge=lfs -text
|
||||
*.gz filter=lfs diff=lfs merge=lfs -text
|
||||
*.h5 filter=lfs diff=lfs merge=lfs -text
|
||||
*.joblib filter=lfs diff=lfs merge=lfs -text
|
||||
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
||||
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
||||
*.model filter=lfs diff=lfs merge=lfs -text
|
||||
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
||||
*.npy filter=lfs diff=lfs merge=lfs -text
|
||||
*.npz filter=lfs diff=lfs merge=lfs -text
|
||||
*.onnx filter=lfs diff=lfs merge=lfs -text
|
||||
*.ot filter=lfs diff=lfs merge=lfs -text
|
||||
*.parquet filter=lfs diff=lfs merge=lfs -text
|
||||
*.pb filter=lfs diff=lfs merge=lfs -text
|
||||
*.pickle filter=lfs diff=lfs merge=lfs -text
|
||||
*.pkl filter=lfs diff=lfs merge=lfs -text
|
||||
*.pt filter=lfs diff=lfs merge=lfs -text
|
||||
*.pth filter=lfs diff=lfs merge=lfs -text
|
||||
*.rar filter=lfs diff=lfs merge=lfs -text
|
||||
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
||||
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
||||
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
||||
*.tar filter=lfs diff=lfs merge=lfs -text
|
||||
*.tflite filter=lfs diff=lfs merge=lfs -text
|
||||
*.tgz filter=lfs diff=lfs merge=lfs -text
|
||||
*.wasm filter=lfs diff=lfs merge=lfs -text
|
||||
*.xz filter=lfs diff=lfs merge=lfs -text
|
||||
*.zip filter=lfs diff=lfs merge=lfs -text
|
||||
*.zst filter=lfs diff=lfs merge=lfs -text
|
||||
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
||||
tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
||||
93
README.md
Normal file
93
README.md
Normal file
@@ -0,0 +1,93 @@
|
||||
---
|
||||
license: apache-2.0
|
||||
base_model: Qwen/Qwen2.5-0.5B
|
||||
pipeline_tag: text-generation
|
||||
tags:
|
||||
- authorization
|
||||
- ai-agent-safety
|
||||
- verifier-as-reward
|
||||
- rlvr
|
||||
---
|
||||
|
||||
# verifier-ce-qwen2.5-0.5b
|
||||
|
||||
A **Qwen2.5-0.5B** model fine-tuned to decide **AUTHORIZED / UNAUTHORIZED**
|
||||
for AI-agent actions over an object-capability delegation chain, using a
|
||||
**deterministic authorization verifier as the sole training signal**
|
||||
(verifier cross-entropy — the verdict as a target).
|
||||
|
||||
Companion model to the paper *"Learnable Authorization: A Verifier-as-Reward
|
||||
Benchmark and Method for AI-Agent Authority."* Code, benchmark, and full
|
||||
experiment log: https://github.com/esmaeil-abedi-dev/verifier-as-reward
|
||||
|
||||
## What it does
|
||||
|
||||
Given a natural-language description of a root authority, a chain of
|
||||
delegations (each of which may only narrow the authority it received, with
|
||||
issue/expiry/revocation times), and a pending action, the model answers
|
||||
whether the action is authorized. The training target for every example is
|
||||
the verdict of the deterministic verifier `verify(...)` — never a
|
||||
hand-assigned or stored label.
|
||||
|
||||
## Training
|
||||
|
||||
- **Objective:** verifier cross-entropy, `-log π(verdict)` (see the paper /
|
||||
repo for why sampled REINFORCE and exact expected-reward policy gradient
|
||||
collapse or oscillate at this scale, and why the cross-entropy gradient's
|
||||
non-saturation is what converges).
|
||||
- **Data:** synthetic execution traces across 5 domains and 9 scenario
|
||||
classes, generated and labeled by the verifier; leakage-guarded train /
|
||||
validation / test splits.
|
||||
- **Recipe:** class-balanced reward weighting, natural-language prompts,
|
||||
full-parameter fine-tuning.
|
||||
|
||||
## Evaluation
|
||||
|
||||
Committed held-out test set (`benchmark_test.jsonl`, 80 actions,
|
||||
natural-language prompts, evaluated once). Three training seeds:
|
||||
|
||||
| seed | accuracy | false-authorize (headline) | false-refuse |
|
||||
|---|---|---|---|
|
||||
| 7 | 0.988 | 0.023 | 0.000 |
|
||||
| 8 | 0.988 | 0.023 | 0.000 |
|
||||
| **9 (released weights)** | **0.975** | **0.045** | **0.000** |
|
||||
| mean | 0.983 ± 0.006 | 0.030 ± 0.011 | 0.000 |
|
||||
|
||||
The **weights in this repository are the seed-9 checkpoint** (committed-test
|
||||
accuracy 0.975; seeds 7/8 reached 0.988). For calibration, on the identical
|
||||
prompts and metric: claude-sonnet-4.5 0.975, deepseek-r1 0.95,
|
||||
gemini-2.5-flash 0.875, a lexical heuristic floor 0.80, llama-3.3-70b 0.775,
|
||||
untrained baseline ~0.47. The only imperfect class is `chain_structure`
|
||||
(0.90) — the hardest structural-reasoning class.
|
||||
|
||||
**These are in-distribution results** (train/validation/test drawn from one
|
||||
generator, deduplicated at the decision-context level). They do not by
|
||||
themselves establish out-of-distribution generalization; see the
|
||||
repository's domain-hold-out experiment.
|
||||
|
||||
## Intended use and limitations
|
||||
|
||||
- **Research artifact**, not a production authorization system. The verifier
|
||||
itself is the ground truth; this model approximates it.
|
||||
- Trained and evaluated on a **synthetic** benchmark from a single verifier;
|
||||
generalization to real agent logs or to authorization models outside the
|
||||
attenuated-delegation formalism is not established here.
|
||||
- See the repository's `EXPERIMENT_LOG.md` for exact metrics (including the
|
||||
in-distribution vs. out-of-distribution domain-hold-out results) and their
|
||||
confidence intervals, and `DATASHEET.md` for the benchmark's limitations.
|
||||
|
||||
## Usage
|
||||
|
||||
```python
|
||||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||||
tok = AutoTokenizer.from_pretrained("esmaeil-abedi-dev/verifier-ce-qwen2.5-0.5b")
|
||||
model = AutoModelForCausalLM.from_pretrained("esmaeil-abedi-dev/verifier-ce-qwen2.5-0.5b")
|
||||
```
|
||||
|
||||
Or evaluate it on any benchmark split with the repo's harness (no retraining):
|
||||
|
||||
```bash
|
||||
PYTHONPATH=. python train_verifier_reward.py \
|
||||
--eval-checkpoint esmaeil-abedi-dev/verifier-ce-qwen2.5-0.5b \
|
||||
--test-file benchmark_test.jsonl
|
||||
```
|
||||
54
chat_template.jinja
Normal file
54
chat_template.jinja
Normal file
@@ -0,0 +1,54 @@
|
||||
{%- if tools %}
|
||||
{{- '<|im_start|>system\n' }}
|
||||
{%- if messages[0]['role'] == 'system' %}
|
||||
{{- messages[0]['content'] }}
|
||||
{%- else %}
|
||||
{{- 'You are a helpful assistant.' }}
|
||||
{%- endif %}
|
||||
{{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
||||
{%- for tool in tools %}
|
||||
{{- "\n" }}
|
||||
{{- tool | tojson }}
|
||||
{%- endfor %}
|
||||
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
||||
{%- else %}
|
||||
{%- if messages[0]['role'] == 'system' %}
|
||||
{{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- for message in messages %}
|
||||
{%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
|
||||
{%- elif message.role == "assistant" %}
|
||||
{{- '<|im_start|>' + message.role }}
|
||||
{%- if message.content %}
|
||||
{{- '\n' + message.content }}
|
||||
{%- endif %}
|
||||
{%- for tool_call in message.tool_calls %}
|
||||
{%- if tool_call.function is defined %}
|
||||
{%- set tool_call = tool_call.function %}
|
||||
{%- endif %}
|
||||
{{- '\n<tool_call>\n{"name": "' }}
|
||||
{{- tool_call.name }}
|
||||
{{- '", "arguments": ' }}
|
||||
{{- tool_call.arguments | tojson }}
|
||||
{{- '}\n</tool_call>' }}
|
||||
{%- endfor %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- elif message.role == "tool" %}
|
||||
{%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
|
||||
{{- '<|im_start|>user' }}
|
||||
{%- endif %}
|
||||
{{- '\n<tool_response>\n' }}
|
||||
{{- message.content }}
|
||||
{{- '\n</tool_response>' }}
|
||||
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- endfor %}
|
||||
{%- if add_generation_prompt %}
|
||||
{{- '<|im_start|>assistant\n' }}
|
||||
{%- endif %}
|
||||
58
config.json
Normal file
58
config.json
Normal file
@@ -0,0 +1,58 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen2ForCausalLM"
|
||||
],
|
||||
"attention_dropout": 0.0,
|
||||
"bos_token_id": 151643,
|
||||
"dtype": "float16",
|
||||
"eos_token_id": 151643,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 896,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 4864,
|
||||
"layer_types": [
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention"
|
||||
],
|
||||
"max_position_embeddings": 32768,
|
||||
"max_window_layers": 24,
|
||||
"model_type": "qwen2",
|
||||
"num_attention_heads": 14,
|
||||
"num_hidden_layers": 24,
|
||||
"num_key_value_heads": 2,
|
||||
"pad_token_id": null,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_parameters": {
|
||||
"rope_theta": 1000000.0,
|
||||
"rope_type": "default"
|
||||
},
|
||||
"sliding_window": null,
|
||||
"tie_word_embeddings": true,
|
||||
"transformers_version": "5.13.1",
|
||||
"use_cache": true,
|
||||
"use_mrope": false,
|
||||
"use_sliding_window": false,
|
||||
"vocab_size": 151936
|
||||
}
|
||||
7
generation_config.json
Normal file
7
generation_config.json
Normal file
@@ -0,0 +1,7 @@
|
||||
{
|
||||
"bos_token_id": 151643,
|
||||
"do_sample": false,
|
||||
"eos_token_id": 151643,
|
||||
"max_new_tokens": 2048,
|
||||
"transformers_version": "5.13.1"
|
||||
}
|
||||
3
model.safetensors
Normal file
3
model.safetensors
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:eefae33fd8974308a190a03b3329e09b8b0327f77aff780a339588ddc7174612
|
||||
size 988097536
|
||||
3
tokenizer.json
Normal file
3
tokenizer.json
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:3fd169731d2cbde95e10bf356d66d5997fd885dd8dbb6fb4684da3f23b2585d8
|
||||
size 11421892
|
||||
30
tokenizer_config.json
Normal file
30
tokenizer_config.json
Normal file
@@ -0,0 +1,30 @@
|
||||
{
|
||||
"add_prefix_space": false,
|
||||
"backend": "tokenizers",
|
||||
"bos_token": null,
|
||||
"clean_up_tokenization_spaces": false,
|
||||
"eos_token": "<|endoftext|>",
|
||||
"errors": "replace",
|
||||
"extra_special_tokens": [
|
||||
"<|im_start|>",
|
||||
"<|im_end|>",
|
||||
"<|object_ref_start|>",
|
||||
"<|object_ref_end|>",
|
||||
"<|box_start|>",
|
||||
"<|box_end|>",
|
||||
"<|quad_start|>",
|
||||
"<|quad_end|>",
|
||||
"<|vision_start|>",
|
||||
"<|vision_end|>",
|
||||
"<|vision_pad|>",
|
||||
"<|image_pad|>",
|
||||
"<|video_pad|>"
|
||||
],
|
||||
"is_local": true,
|
||||
"local_files_only": false,
|
||||
"model_max_length": 131072,
|
||||
"pad_token": "<|endoftext|>",
|
||||
"split_special_tokens": false,
|
||||
"tokenizer_class": "Qwen2Tokenizer",
|
||||
"unk_token": null
|
||||
}
|
||||
Reference in New Issue
Block a user