Report every pass@k from 1 through 16

crosscodeeval-csharp-v5/pass_counts.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7268f9332c4f44d4666484859eab1ee98ec0b8c7a2edf316d179e2ac055a057a
-size 204815
+oid sha256:4f0c566d17837c7a9628bbe7f9097a32ef46ef24330db4993bd84edaf3823ee7
+size 213144

crosscodeeval-java-v4/pass_counts.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d628e6d8e1ffeab379540180d596f203f0790fd0a8356f28b270286633c8fbcd
-size 295575
+oid sha256:ee9e0a1a183762facbd2604098c14c606e40afa16b538e6c58588b323e26a0a1
+size 310701

crosscodeeval-python-v3/pass_counts.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a24c00b53750e260ad9b408a1a179303c8f15818643fc95f08084eb74fcdbc28
-size 69058
+oid sha256:6e6d1a167d1c8e73957f34e2eb1ae22b803c5641fd8c02937721dbd8a914d297
+size 75110

crosscodeeval-typescript-v3/pass_counts.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:149598fdce4fe428abf02ff539fa6e3fe73e4465d43cba3c44b7e86f1d685c1a
-size 466101
+oid sha256:e33626110de802f7ed9a683805eeb0492ed3d6ce480675934ee37622d67a4a28
+size 480717

runs/2026-10-07-crosscodeeval-pass16.json

@@ -288,7 +288,18 @@
           }
         },
         "trajectory_files_found": 107359
-      }
+      },
+      "pass@3": 0.05105306578667235,
+      "pass@5": 0.06333516210565392,
+      "pass@6": 0.0679098658420566,
+      "pass@7": 0.07186940481277292,
+      "pass@9": 0.07855124383812909,
+      "pass@10": 0.08145727223596076,
+      "pass@11": 0.08415129952015198,
+      "pass@12": 0.08667226216406544,
+      "pass@13": 0.0890496593570364,
+      "pass@14": 0.09130650769995032,
+      "pass@15": 0.09346125186289121
     },
     "qwen3-30b-thinking": {
       "tasks": 6710,
@@ -481,7 +492,18 @@
           }
         },
         "trajectory_files_found": 107360
-      }
+      },
+      "pass@3": 0.049585373642750695,
+      "pass@5": 0.06213431377365804,
+      "pass@6": 0.06684295585189773,
+      "pass@7": 0.07090064927620815,
+      "pass@9": 0.07763596500369974,
+      "pass@10": 0.08049430078118602,
+      "pass@11": 0.08309433053285512,
+      "pass@12": 0.08547911105288154,
+      "pass@13": 0.08768229721098575,
+      "pass@14": 0.08973050173869847,
+      "pass@15": 0.09164493293591654
     }
   },
   "stage7": "runs/2026-10-07-crosscodeeval-pass16/stage-7-combined.json.gz",
@@ -490,5 +512,23 @@
     "Includes original collection and accepted recovery attempts.",
     "One context-limit failure counts as reward 0 under the agreed evaluation policy; its absent verifier score and original error remain recorded."
   ],
-  "pass_at_k": "1 - comb(16 - solved, k) / comb(16, k)"
+  "pass_at_k": "1 - comb(16 - solved, k) / comb(16, k)",
+  "reported_k": [
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    7,
+    8,
+    9,
+    10,
+    11,
+    12,
+    13,
+    14,
+    15,
+    16
+  ]
 }

runs/2026-10-07-crosscodeeval-pass16/add_all_k.py

@@ -0,0 +1,56 @@
+"""Add every pass@k from 1 through 16 to the existing teacher-results PR."""
+from pathlib import Path
+import sys,json,math,hashlib,shutil
+import pyarrow as pa
+import pyarrow.parquet as pq
+from huggingface_hub import HfApi,CommitOperationAdd,hf_hub_download
+root=Path(__file__).resolve().parent;release=root/'release';run='2026-10-07-crosscodeeval-pass16';out=release/'runs'/run
+sys.path.insert(0,'/home/hpc/y500bb/y500bb12/OT_next_data')
+from validation.checks.reward_metrics import pass_at_k
+from validation.reporting.plot_results import rates
+state=json.loads((root/'hf-publication.json').read_text());changed=[];groups={}
+for p in sorted(release.glob('crosscodeeval-*/pass_counts.parquet')):
+ rows=pq.read_table(p).to_pylist()
+ for row in rows:
+  if row['run']!=f'runs/{run}.json':continue
+  n,c=row['attempts'],row['solved'];assert n==16
+  for k in range(1,17):
+   value=pass_at_k(n,c,k);expected=1-(math.comb(n-c,k)/math.comb(n,k) if n-c>=k else 0)
+   assert abs(value-expected)<1e-12
+   row[f'pass@{k}']=value
+  assert all(row[f'pass@{k}']<=row[f'pass@{k+1}'] for k in range(1,16))
+ pq.write_table(pa.Table.from_pylist(rows),p);changed.append(p)
+summary=json.loads((out/'summary.json').read_text())
+for model in summary:
+ grouped=rates(out/f'{model}-stage-6.json',list(range(1,17)));groups[model]=grouped
+ summary[model].update({f'pass@{k}':v for k,v in grouped['All tasks'].items()})
+(out/'summary.json').write_text(json.dumps(summary,indent=2)+'\n');changed.append(out/'summary.json')
+p=release/'runs'/f'{run}.json';record=json.loads(p.read_text());record['summary']=summary;record['reported_k']=list(range(1,17));p.write_text(json.dumps(record,indent=2)+'\n');changed.append(p)
+p=out/'pass-at-k.json';p.write_text(json.dumps(groups,indent=2)+'\n');changed.append(p)
+import matplotlib
+matplotlib.use('Agg')
+import matplotlib.pyplot as plt
+labels=['csharp','java','python','typescript','All tasks'];fig,axes=plt.subplots(1,2,figsize=(13,4.8),sharey=True,layout='constrained')
+for ax,(model,grouped) in zip(axes,groups.items()):
+ for label in labels:
+  values=grouped[label];ax.plot(list(values),[100*v for v in values.values()],marker='o',markersize=3,label=label,linewidth=2 if label=='All tasks' else 1.3)
+ ax.set(title='Qwen Instruct' if 'instruct' in model else 'Qwen Thinking',xlabel='k (attempts)',xticks=range(1,17),xlim=(1,16),ylim=(0,20));ax.grid(alpha=.2);ax.spines[['top','right']].set_visible(False);ax.legend(frameon=False)
+axes[0].set_ylabel('Pass@k (%)')
+for ext in ['png','svg']:
+ p=out/f'pass-rates-all-k.{ext}';fig.savefig(p,dpi=180);changed.append(p)
+plt.close(fig)
+# Keep the original preparation script consistent for per-task and summary exports.
+p=root/'prepare.py';s=p.read_text().replace('for k in [1,2,4,8,16]','for k in range(1,17)');p.write_text(s);shutil.copyfile(p,out/p.name);changed.append(out/p.name)
+shutil.copyfile(Path(__file__),out/Path(__file__).name);changed.append(out/Path(__file__).name)
+p=root/'pull-request.md';body=p.read_text().replace('pass@1/2/4/8/16','pass@k for every k from 1 through 16').replace('Plots show pass rates by language,','Plots show pass rates for every k from 1 through 16, by language,');p.write_text(body)
+api=HfApi();commit=api.create_commit(repo_id=state['repo'],repo_type='dataset',revision=state['revision'],parent_commit=state['commit'],operations=[CommitOperationAdd(path_in_repo=str(p.relative_to(release)),path_or_fileobj=str(p)) for p in changed],commit_message='Report every pass@k from 1 through 16')
+state['commit']=commit.oid;state['status']='all-k-uploaded';(root/'hf-publication.json').write_text(json.dumps(state,indent=2)+'\n')
+for p in changed:
+ remote=hf_hub_download(state['repo'],str(p.relative_to(release)),repo_type='dataset',revision=commit.oid,cache_dir=str(root/'verify-cache'))
+ assert hashlib.sha256(Path(remote).read_bytes()).digest()==hashlib.sha256(p.read_bytes()).digest()
+# The opening discussion comment contains the PR description.
+d=api.get_discussion_details(state['repo'],6,repo_type='dataset')
+comments=[e for e in d.events if e.type=='comment']
+original=next(e for e in comments if 'Adds per-task teacher results' in e.content)
+api.edit_discussion_comment(repo_id=state['repo'],discussion_num=6,comment_id=original.id,content=body,repo_type='dataset')
+state.update(status='verified',reported_k=list(range(1,17)),files=sum(p.is_file() for p in release.rglob('*')));(root/'hf-publication.json').write_text(json.dumps(state,indent=2)+'\n');print(state['pr_url'],flush=True)

runs/2026-10-07-crosscodeeval-pass16/pass-at-k.json

@@ -0,0 +1,186 @@
+{
+  "qwen3-30b-instruct-2507": {
+    "csharp": {
+      "1": 0.011917960088691795,
+      "2": 0.019419807834441978,
+      "3": 0.02475847323408299,
+      "4": 0.0289060532962972,
+      "5": 0.03232556738654299,
+      "6": 0.03526397844579663,
+      "7": 0.037865349727877445,
+      "8": 0.04021940939901029,
+      "9": 0.04238534931439588,
+      "10": 0.044404320734697675,
+      "11": 0.046306166732996,
+      "12": 0.048113268844976166,
+      "13": 0.049842941611234294,
+      "14": 0.0515089923626509,
+      "15": 0.05312269031781227,
+      "16": 0.05469327420546933
+    },
+    "java": {
+      "1": 0.05448548812664908,
+      "2": 0.07895338610378189,
+      "3": 0.09449962306822465,
+      "4": 0.10587984574791963,
+      "5": 0.11489097973266839,
+      "6": 0.12238421209397463,
+      "7": 0.12882714910419396,
+      "8": 0.13450297576946652,
+      "9": 0.13959522667306307,
+      "10": 0.14422931949317175,
+      "11": 0.14849517237380033,
+      "12": 0.15246020470294874,
+      "13": 0.15617696946852622,
+      "14": 0.1596877748460862,
+      "15": 0.16302770448548812,
+      "16": 0.1662269129287599
+    },
+    "python": {
+      "1": 0.03125,
+      "2": 0.0484375,
+      "3": 0.05991236772486772,
+      "4": 0.06853632478632478,
+      "5": 0.07545723951973952,
+      "6": 0.08125844063344063,
+      "7": 0.08629261363636365,
+      "8": 0.09078912342801232,
+      "9": 0.09489971833721834,
+      "10": 0.09872419247419248,
+      "11": 0.10232721560846561,
+      "12": 0.10575015262515262,
+      "13": 0.10901951058201058,
+      "14": 0.11215277777777778,
+      "15": 0.11516203703703703,
+      "16": 0.11805555555555555
+    },
+    "typescript": {
+      "1": 0.01808993399339934,
+      "2": 0.027816281628162815,
+      "3": 0.03435938236680811,
+      "4": 0.03926921263554927,
+      "5": 0.043210069633336956,
+      "6": 0.04651528339647151,
+      "7": 0.04937492787740312,
+      "8": 0.05190762666010191,
+      "9": 0.054192024971727946,
+      "10": 0.05628245516859379,
+      "11": 0.05821757999975821,
+      "12": 0.060025568490915025,
+      "13": 0.061727369165487975,
+      "14": 0.06333883388338833,
+      "15": 0.06487211221122112,
+      "16": 0.06633663366336634
+    },
+    "All tasks": {
+      "1": 0.027971311475409835,
+      "2": 0.04189269746646796,
+      "3": 0.05105306578667235,
+      "4": 0.05787564894122272,
+      "5": 0.06333516210565392,
+      "6": 0.0679098658420566,
+      "7": 0.07186940481277292,
+      "8": 0.07538018034292251,
+      "9": 0.07855124383812909,
+      "10": 0.08145727223596076,
+      "11": 0.08415129952015198,
+      "12": 0.08667226216406544,
+      "13": 0.0890496593570364,
+      "14": 0.09130650769995032,
+      "15": 0.09346125186289121,
+      "16": 0.09552906110283159
+    }
+  },
+  "qwen3-30b-thinking": {
+    "csharp": {
+      "1": 0.009931633407243163,
+      "2": 0.01701157920670116,
+      "3": 0.02255437651779115,
+      "4": 0.027151303980572274,
+      "5": 0.031093229263960975,
+      "6": 0.03454140094938321,
+      "7": 0.037596909742142554,
+      "8": 0.040330417714009734,
+      "9": 0.04279572976912223,
+      "10": 0.04503626454845967,
+      "11": 0.0470882437345852,
+      "12": 0.04898231849451362,
+      "13": 0.05074437757364587,
+      "14": 0.05239591032273959,
+      "15": 0.05395417590539542,
+      "16": 0.05543237250554324
+    },
+    "java": {
+      "1": 0.053067282321899736,
+      "2": 0.0768117854001759,
+      "3": 0.09180738786279682,
+      "4": 0.10255617733190292,
+      "5": 0.11089852320063402,
+      "6": 0.11771994496268902,
+      "7": 0.1234975644408362,
+      "8": 0.12851186925065552,
+      "9": 0.13294079930623465,
+      "10": 0.1369052451506277,
+      "11": 0.1404922584012294,
+      "12": 0.1437672881208501,
+      "13": 0.14678100263852242,
+      "14": 0.14957343887423044,
+      "15": 0.1521767810026385,
+      "16": 0.1546174142480211
+    },
+    "python": {
+      "1": 0.025607638888888888,
+      "2": 0.04157021604938271,
+      "3": 0.053290343915343916,
+      "4": 0.06278998778998779,
+      "5": 0.07087265550807217,
+      "6": 0.07792525761275762,
+      "7": 0.08418540371665371,
+      "8": 0.08982290857290857,
+      "9": 0.09496527777777777,
+      "10": 0.09970845127095128,
+      "11": 0.10412374084249085,
+      "12": 0.10826338013838013,
+      "13": 0.11216517857142858,
+      "14": 0.11585648148148148,
+      "15": 0.1193576388888889,
+      "16": 0.12268518518518519
+    },
+    "typescript": {
+      "1": 0.018234323432343234,
+      "2": 0.027967546754675464,
+      "3": 0.03472123998114097,
+      "4": 0.03996391397381496,
+      "5": 0.04425162296449425,
+      "6": 0.04786756148142286,
+      "7": 0.05098308869348473,
+      "8": 0.053712037870453706,
+      "9": 0.05613428650557363,
+      "10": 0.05830798739214581,
+      "11": 0.06027665953408528,
+      "12": 0.06207358647952707,
+      "13": 0.06372465818010373,
+      "14": 0.06525027502750275,
+      "15": 0.06666666666666667,
+      "16": 0.06798679867986798
+    },
+    "All tasks": {
+      "1": 0.02687220566318927,
+      "2": 0.0404284649776453,
+      "3": 0.049585373642750695,
+      "4": 0.05652691570724358,
+      "5": 0.06213431377365804,
+      "6": 0.06684295585189773,
+      "7": 0.07090064927620815,
+      "8": 0.07446319204888505,
+      "9": 0.07763596500369974,
+      "10": 0.08049430078118602,
+      "11": 0.08309433053285512,
+      "12": 0.08547911105288154,
+      "13": 0.08768229721098575,
+      "14": 0.08973050173869847,
+      "15": 0.09164493293591654,
+      "16": 0.09344262295081968
+    }
+  }
+}

runs/2026-10-07-crosscodeeval-pass16/prepare.py

@@ -110,12 +110,12 @@ for model,rows in allrows.items():
  mine=[r for r in final if r['model']==model];items=[]
  for r in mine:
   assert len(r['rewards'])==16 and r['complete'];n=int(r['solved']);language=r['task_id'].split('-')[1]
-  record={'path':r['task_id'],'content_sha256':r['sha256'],'model':model,'attempts':16,'graded':16,'solved':n,'group':reward_group(r['rewards']),'run':f'runs/{run}.json',**{f'pass@{k}':pass_at_k(16,n,k) for k in [1,2,4,8,16]}}
+  record={'path':r['task_id'],'content_sha256':r['sha256'],'model':model,'attempts':16,'graded':16,'solved':n,'group':reward_group(r['rewards']),'run':f'runs/{run}.json',**{f'pass@{k}':pass_at_k(16,n,k) for k in range(1,17)}}
   tables[folders[language]].append(record)
   items.append({'task':r['task_id'],'rewards':r['rewards'],'metrics':{'attempts':16,'solved':n,'graded':16}})
  stage6=out/f'{model}-stage-6.json';write(stage6,{'stage':6,'complete':True,'dry_run':False,'model':model,'items':items})
  panels.append((model,rates(stage6,[1,2,4,8,16])));plot_solved_counts(solved_counts(stage6),out/f'{model}-solved-counts.png')
- summary[model]={'tasks':len(mine),'attempts':len(rows),**{f'pass@{k}':sum(pass_at_k(16,int(r['solved']),k) for r in mine)/len(mine) for k in [1,2,4,8,16]},'trace_metrics':a}
+ summary[model]={'tasks':len(mine),'attempts':len(rows),**{f'pass@{k}':sum(pass_at_k(16,int(r['solved']),k) for r in mine)/len(mine) for k in range(1,17)},'trace_metrics':a}
 plot_rates(panels,[1,2,4,8,16],out/'pass-rates.png')
 zipped(out/'stage-7-combined.json.gz',combined)
 write(out/'summary.json',summary)

runs/2026-10-07-crosscodeeval-pass16/summary.json

@@ -285,7 +285,18 @@
         }
       },
       "trajectory_files_found": 107359
-    }
+    },
+    "pass@3": 0.05105306578667235,
+    "pass@5": 0.06333516210565392,
+    "pass@6": 0.0679098658420566,
+    "pass@7": 0.07186940481277292,
+    "pass@9": 0.07855124383812909,
+    "pass@10": 0.08145727223596076,
+    "pass@11": 0.08415129952015198,
+    "pass@12": 0.08667226216406544,
+    "pass@13": 0.0890496593570364,
+    "pass@14": 0.09130650769995032,
+    "pass@15": 0.09346125186289121
   },
   "qwen3-30b-thinking": {
     "tasks": 6710,
@@ -478,6 +489,17 @@
         }
       },
       "trajectory_files_found": 107360
-    }
+    },
+    "pass@3": 0.049585373642750695,
+    "pass@5": 0.06213431377365804,
+    "pass@6": 0.06684295585189773,
+    "pass@7": 0.07090064927620815,
+    "pass@9": 0.07763596500369974,
+    "pass@10": 0.08049430078118602,
+    "pass@11": 0.08309433053285512,
+    "pass@12": 0.08547911105288154,
+    "pass@13": 0.08768229721098575,
+    "pass@14": 0.08973050173869847,
+    "pass@15": 0.09164493293591654
   }
 }