Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77281.72342552035,
-      "required_pages_per_second_upper_estimate": 0.005346528944881086
+      "required_token_positions_per_second": 77302.1223930104,
+      "required_pages_per_second_upper_estimate": 0.005347940192784174
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9660.215428190044,
-      "required_pages_per_second_upper_estimate": 0.004044530471545748
+      "required_token_positions_per_second": 9662.7652991263,
+      "required_pages_per_second_upper_estimate": 0.004045598049259399
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19320.430856380088,
-      "required_pages_per_second_upper_estimate": 0.004023936233358955
+      "required_token_positions_per_second": 19325.5305982526,
+      "required_pages_per_second_upper_estimate": 0.004024998375101762
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28980.64628457013,
-      "required_pages_per_second_upper_estimate": 0.0017373454938753121
+      "required_token_positions_per_second": 28988.2958973789,
+      "required_pages_per_second_upper_estimate": 0.0017378040764829154
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28980.64628457013,
-      "required_pages_per_second_upper_estimate": 0.0022531177762340196
+      "required_token_positions_per_second": 28988.2958973789,
+      "required_pages_per_second_upper_estimate": 0.0022537124999828105
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28980.64628457013,
-      "required_pages_per_second_upper_estimate": 0.37173261354485104
+      "required_token_positions_per_second": 28988.2958973789,
+      "required_pages_per_second_upper_estimate": 0.3718307345644476
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10943998497663188,
-    "producer_p95_seconds": 0.31324307200702606,
-    "gpu_update_seconds": 0.9540159915177717,
-    "producer_batches_per_second": 9.137428154924587,
-    "gpu_batches_per_second": 1.048200458787982,
-    "producer_headroom": 8.717252581142784,
+    "producer_batch_seconds": 0.12196040357042648,
+    "producer_p95_seconds": 0.3336775379939354,
+    "gpu_update_seconds": 0.9537642398116928,
+    "producer_batches_per_second": 8.199382510427217,
+    "gpu_batches_per_second": 1.048477137492003,
+    "producer_headroom": 7.820277826982904,
     "sustainable": true,
-    "buffer_seconds": 60.10300746561962,
+    "buffer_seconds": 60.08714710813665,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405045.0908498902,
+    "generation_bytes_per_second": 404920.6724565852,
     "upload_bytes_per_second": 5596837.472966093,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.9132475292447,
+    "estimated_full_shards_per_hour": 86.88655023835342,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 723200, "loss": 0.5436961054801941, "lr": 8e-05, "grad_norm": 0.22530942146442387, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105856894818, "tokens_per_second": 151856.9309731779, "data_wait_s": 1.3899996702093631e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723300, "loss": 0.5442550778388977, "lr": 8e-05, "grad_norm": 0.21169117058640777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105871498737, "tokens_per_second": 152329.35673679033, "data_wait_s": 1.446500391466543e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723400, "loss": 0.5364526510238647, "lr": 8e-05, "grad_norm": 0.2332380704781859, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105885999701, "tokens_per_second": 151139.34921965763, "data_wait_s": 1.1629002983681858e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 723500, "loss": 0.5505629181861877, "lr": 8e-05, "grad_norm": 0.23283509450779433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105900622546, "tokens_per_second": 152449.045320536, "data_wait_s": 1.4262004697229713e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6c19a663097130a9731b3eebf42e96a4c4b75999b6bfe55145df4e262099d21e
-size 818958
+oid sha256:014ce36cf6017ea4a2bf23344970db8cd6e2861ddecbac5bb4484d8edbd8ba69
+size 821076