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": 76668.18917321526,
-      "required_pages_per_second_upper_estimate": 0.005304083221711021
+      "required_token_positions_per_second": 77752.29153406943,
+      "required_pages_per_second_upper_estimate": 0.005379083938498954
     },
     {
       "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": 9583.523646651907,
-      "required_pages_per_second_upper_estimate": 0.004012421224122242
+      "required_token_positions_per_second": 9719.036441758679,
+      "required_pages_per_second_upper_estimate": 0.004069157601604495
     },
     {
       "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": 19167.047293303815,
-      "required_pages_per_second_upper_estimate": 0.0039919904821667405
+      "required_token_positions_per_second": 19438.072883517358,
+      "required_pages_per_second_upper_estimate": 0.004048437965182764
     },
     {
       "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": 28750.57093995572,
-      "required_pages_per_second_upper_estimate": 0.0017235528282703882
+      "required_token_positions_per_second": 29157.10932527604,
+      "required_pages_per_second_upper_estimate": 0.0017479241837221695
     },
     {
       "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": 28750.57093995572,
-      "required_pages_per_second_upper_estimate": 0.002235230430184739
+      "required_token_positions_per_second": 29157.10932527604,
+      "required_pages_per_second_upper_estimate": 0.002266837001470017
     },
     {
       "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": 28750.57093995572,
-      "required_pages_per_second_upper_estimate": 0.3687814540597955
+      "required_token_positions_per_second": 29157.10932527604,
+      "required_pages_per_second_upper_estimate": 0.3739960919597753
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08411015249996012,
-    "producer_p95_seconds": 0.21980245200029458,
-    "gpu_update_seconds": 0.9616504680112826,
-    "producer_batches_per_second": 11.88917116754097,
-    "gpu_batches_per_second": 1.039878867909278,
-    "producer_headroom": 11.433227017532023,
+    "producer_batch_seconds": 0.08700810922641722,
+    "producer_p95_seconds": 0.198406937000982,
+    "gpu_update_seconds": 0.9482421488207061,
+    "producer_batches_per_second": 11.493181599863822,
+    "gpu_batches_per_second": 1.0545829472394401,
+    "producer_headroom": 10.898319217041472,
     "sustainable": true,
-    "buffer_seconds": 60.5839794847108,
+    "buffer_seconds": 60.68749752452519,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334706.06432043517,
-    "upload_bytes_per_second": 5549068.002570031,
+    "generation_bytes_per_second": 335400.0231646085,
+    "upload_bytes_per_second": 5443089.413309285,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.82012984475891,
+    "estimated_full_shards_per_hour": 71.9690372581834,
     "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": [
-    {
-      "seconds": 1.1562814480002999,
-      "bytes": 11922123,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.2267546230013977,
       "bytes": 11778233,
@@ -562,6 +556,12 @@
       "bytes": 6152237,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8757686259996262,
+      "bytes": 2112854,
+      "pages": 23,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -172,3 +172,4 @@
 {"step": 542200, "loss": 0.5227675437927246, "lr": 0.0003, "grad_norm": 0.1815667588109992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79360273684, "tokens_per_second": 151991.64493333848, "data_wait_s": 1.5243000234477222e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 542300, "loss": 0.5763422846794128, "lr": 0.0003, "grad_norm": 0.1842850120887658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79374885018, "tokens_per_second": 152393.62401247176, "data_wait_s": 7.462000212399289e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 542400, "loss": 0.5786153674125671, "lr": 0.0003, "grad_norm": 0.1579134351227863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79389561267, "tokens_per_second": 154388.78592248357, "data_wait_s": 1.460000203223899e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 542500, "loss": 0.6364598870277405, "lr": 0.0003, "grad_norm": 0.16701580214921663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79404247980, "tokens_per_second": 152739.05460617493, "data_wait_s": 8.732997230254114e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:96f2472d6e075a2cb0eaa662bf450479cf53c80507f772af36f4eec001b17d3b
-size 371046
+oid sha256:6a660b13c05a6740e7eebdf0a5344e00e6624d9749840bfb591f869aea01b3cf
+size 373164