Update 2B adapter with repair-focused run

README.md

@@ -1,41 +1,62 @@
 ---
 base_model: Qwen/Qwen3.5-2B
 library_name: peft
+model_name: model
 tags:
+- base_model:adapter:Qwen/Qwen3.5-2B
 - lora
-- qwen
-- bash
-- shell
-- linux
-- text-generation
+- sft
+- transformers
+- trl
+licence: license
+pipeline_tag: text-generation
 ---
 
-# Qwen3.5-2B ShellCommand-Linux LoRA
+# Model Card for model
 
-This repository contains a PEFT LoRA adapter trained for Linux natural-language-to-shell translation.
+This model is a fine-tuned version of [Qwen/Qwen3.5-2B](https://huggingface.co/Qwen/Qwen3.5-2B).
+It has been trained using [TRL](https://github.com/huggingface/trl).
 
-## Artifact Type
+## Quick start
 
-This is a LoRA adapter, not a merged full model checkpoint.
+```python
+from transformers import pipeline
 
-## Intended Behavior
+question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
+generator = pipeline("text-generation", model="None", device="cuda")
+output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
+print(output["generated_text"])
+```
 
-The model is tuned to return exactly one of:
+## Training procedure
 
-- a Bash command or short Bash snippet
-- `ASK: <one short clarifying question>`
-- `CANNOT: <brief reason>`
+ 
 
-## Eval Snapshot
 
-- score: `218.4816`
-- verifier ok rate: `61.67%`
-- verifier command rate: `59.38%`
-- verifier ask rate: `56.25%`
-- verifier cannot rate: `100.00%`
-- exact any-exact rate: `24.00%`
-- exact parse-ok rate: `98.00%`
 
-## Usage
+This model was trained with SFT.
 
-Load this adapter on top of `Qwen/Qwen3.5-2B` with PEFT.
+### Framework versions
+
+- PEFT 0.18.1
+- TRL: 0.29.0
+- Transformers: 5.3.0
+- Pytorch: 2.10.0
+- Datasets: 4.8.2
+- Tokenizers: 0.22.2
+
+## Citations
+
+
+
+Cite TRL as:
+    
+```bibtex
+@software{vonwerra2020trl,
+  title   = {{TRL: Transformers Reinforcement Learning}},
+  author  = {von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin},
+  license = {Apache-2.0},
+  url     = {https://github.com/huggingface/trl},
+  year    = {2020}
+}
+```

adapter_config.json

@@ -29,13 +29,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
+    "k_proj",
     "gate_proj",
-    "o_proj",
-    "q_proj",
     "v_proj",
-    "up_proj",
     "down_proj",
-    "k_proj"
+    "o_proj",
+    "up_proj",
+    "q_proj"
   ],
   "target_parameters": null,
   "task_type": "CAUSAL_LM",

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:312ea467f4f51a79d21cf2af71f40f8b5f3d75d3a8e5eceb88643a90a01b603b
+oid sha256:da96b3764db1705fad7f4ef8ebf236e0078833e3439c9ea42e37905bd501af44
 size 43672416

eval/exact_summary.json

@@ -1,14 +1,14 @@
 {
-  "model": "/root/bitnet-nl2sh/output/autoresearch_proxy_qwen35_2b/full_v1/qwen35_2b_batch8_ask_full_v1/model",
+  "model": "/root/bitnet-nl2sh/output/autoresearch_proxy_qwen35_2b/repair_full_v1/qwen35_2b_batch8_repair_full_v1/model",
   "base_model": "Qwen/Qwen3.5-2B",
   "test_file": "/root/bitnet-nl2sh/output/data/nl2sh_test_raw.jsonl",
   "rows": 100,
-  "primary_exact": 20,
+  "primary_exact": 11,
   "alt_exact": 6,
-  "any_exact": 24,
-  "parse_ok": 98,
-  "primary_exact_rate": 0.2,
-  "any_exact_rate": 0.24,
-  "parse_ok_rate": 0.98,
-  "avg_gen_seconds_per_example": 0.5884
+  "any_exact": 15,
+  "parse_ok": 97,
+  "primary_exact_rate": 0.11,
+  "any_exact_rate": 0.15,
+  "parse_ok_rate": 0.97,
+  "avg_gen_seconds_per_example": 0.5498
 }

eval/score.json

@@ -1,17 +1,17 @@
 {
   "components": {
-    "any_exact_rate": 0.24,
+    "any_exact_rate": 0.15,
     "ask_rate": 0.5625,
     "cannot_rate": 1.0,
-    "command_rate": 0.5938,
-    "ok_rate": 0.6167,
-    "parse_ok_rate": 0.98,
-    "weak_category_rate": 0.5357142857142857
+    "command_rate": 0.6875,
+    "ok_rate": 0.6917,
+    "parse_ok_rate": 0.97,
+    "weak_category_rate": 0.6071428571428571
   },
-  "exact_summary": "/root/bitnet-nl2sh/output/autoresearch_proxy_qwen35_2b/full_v1/qwen35_2b_batch8_ask_full_v1/eval_exact/summary.json",
-  "score": 218.4816,
-  "score_0_to_1": 2.184816,
-  "verifier_summary": "/root/bitnet-nl2sh/output/autoresearch_proxy_qwen35_2b/full_v1/qwen35_2b_batch8_ask_full_v1/eval_verifier/summary.json",
+  "exact_summary": "/root/bitnet-nl2sh/output/autoresearch_proxy_qwen35_2b/repair_full_v1/qwen35_2b_batch8_repair_full_v1/eval_exact/summary.json",
+  "score": 232.7182,
+  "score_0_to_1": 2.327182,
+  "verifier_summary": "/root/bitnet-nl2sh/output/autoresearch_proxy_qwen35_2b/repair_full_v1/qwen35_2b_batch8_repair_full_v1/eval_verifier/summary.json",
   "weak_categories": [
     "ambiguous_delete",
     "ambiguous_secret",
@@ -25,8 +25,8 @@
     "ambiguous_delete": 0.125,
     "ambiguous_secret": 1.0,
     "create_archive": 1.0,
-    "enabled_services": 0.875,
-    "find_jpgs": 0.0,
+    "enabled_services": 1.0,
+    "find_jpgs": 0.375,
     "json_query": 0.625,
     "top_ips": 0.125
   },

eval/verifier_summary.json

@@ -1,5 +1,5 @@
 {
-  "avg_gen_seconds_per_example": 0.6777,
+  "avg_gen_seconds_per_example": 0.576,
   "base_model": "Qwen/Qwen3.5-2B",
   "category_breakdown": {
     "ambiguous_delete": {
@@ -18,8 +18,8 @@
       "rows": 8
     },
     "count_extension": {
-      "ok": 2,
-      "ok_rate": 0.25,
+      "ok": 0,
+      "ok_rate": 0.0,
       "rows": 8
     },
     "create_archive": {
@@ -33,8 +33,8 @@
       "rows": 8
     },
     "enabled_services": {
-      "ok": 7,
-      "ok_rate": 0.875,
+      "ok": 8,
+      "ok_rate": 1.0,
       "rows": 8
     },
     "extract_archive": {
@@ -43,18 +43,18 @@
       "rows": 8
     },
     "find_jpgs": {
-      "ok": 0,
-      "ok_rate": 0.0,
+      "ok": 3,
+      "ok_rate": 0.375,
       "rows": 8
     },
     "git_branch": {
-      "ok": 0,
-      "ok_rate": 0.0,
+      "ok": 8,
+      "ok_rate": 1.0,
       "rows": 8
     },
     "grep_literal": {
-      "ok": 8,
-      "ok_rate": 1.0,
+      "ok": 4,
+      "ok_rate": 0.5,
       "rows": 8
     },
     "json_query": {
@@ -63,13 +63,13 @@
       "rows": 8
     },
     "replace_literal": {
-      "ok": 2,
-      "ok_rate": 0.25,
+      "ok": 8,
+      "ok_rate": 1.0,
       "rows": 8
     },
     "show_env": {
-      "ok": 8,
-      "ok_rate": 1.0,
+      "ok": 5,
+      "ok_rate": 0.625,
       "rows": 8
     },
     "top_ips": {
@@ -92,14 +92,14 @@
       "rows": 8
     },
     "command": {
-      "ok": 57,
-      "ok_rate": 0.5938,
+      "ok": 66,
+      "ok_rate": 0.6875,
       "rows": 96
     }
   },
-  "model": "/root/bitnet-nl2sh/output/autoresearch_proxy_qwen35_2b/full_v1/qwen35_2b_batch8_ask_full_v1/model",
-  "ok": 74,
-  "ok_rate": 0.6167,
-  "prompt_file": "/root/bitnet-nl2sh/prompts/student_linux_shell_v2f.txt",
+  "model": "/root/bitnet-nl2sh/output/autoresearch_proxy_qwen35_2b/repair_full_v1/qwen35_2b_batch8_repair_full_v1/model",
+  "ok": 83,
+  "ok_rate": 0.6917,
+  "prompt_file": "/root/bitnet-nl2sh/prompts/student_linux_shell_v2g.txt",
   "rows": 120
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5dcc81a623f926adf038f947eeeed6da9a08d33e8cfb9633aeb90d3a34ae660
+oid sha256:90edf4dd4fe7735c98f12ebde98b91799442829e503d750532efa2316597ffa4
 size 5713