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": 77232.82210325013,
-      "required_pages_per_second_upper_estimate": 0.005343145837939732
+      "required_token_positions_per_second": 76856.77133417521,
+      "required_pages_per_second_upper_estimate": 0.005317129773177129
     },
     {
       "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": 9654.102762906266,
-      "required_pages_per_second_upper_estimate": 0.004041971226238397
+      "required_token_positions_per_second": 9607.096416771901,
+      "required_pages_per_second_upper_estimate": 0.004022290650716067
     },
     {
       "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": 19308.205525812533,
-      "required_pages_per_second_upper_estimate": 0.0040213900194053815
+      "required_token_positions_per_second": 19214.192833543802,
+      "required_pages_per_second_upper_estimate": 0.004001809654887225
     },
     {
       "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": 28962.308288718796,
-      "required_pages_per_second_upper_estimate": 0.001736246159024523
+      "required_token_positions_per_second": 28821.289250315705,
+      "required_pages_per_second_upper_estimate": 0.0017277922829958617
     },
     {
       "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": 28962.308288718796,
-      "required_pages_per_second_upper_estimate": 0.0022516920777168976
+      "required_token_positions_per_second": 28821.289250315705,
+      "required_pages_per_second_upper_estimate": 0.002240728467758254
     },
     {
       "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": 28962.308288718796,
-      "required_pages_per_second_upper_estimate": 0.371497393423876
+      "required_token_positions_per_second": 28821.289250315705,
+      "required_pages_per_second_upper_estimate": 0.36968855261368766
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10589470792945122,
-    "producer_p95_seconds": 0.32550206000451,
-    "gpu_update_seconds": 0.9546200435539616,
-    "producer_batches_per_second": 9.44334253857347,
-    "gpu_batches_per_second": 1.0475371921556278,
-    "producer_headroom": 9.014804065467985,
+    "producer_batch_seconds": 0.11047940749978125,
+    "producer_p95_seconds": 0.3204388489975827,
+    "gpu_update_seconds": 0.9592908825096069,
+    "producer_batches_per_second": 9.051460562928707,
+    "gpu_batches_per_second": 1.04243667716709,
+    "producer_headroom": 8.682983591412784,
     "sustainable": true,
-    "buffer_seconds": 60.14106274389958,
+    "buffer_seconds": 60.43532559810524,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448842.637523681,
-    "upload_bytes_per_second": 4989803.359832455,
+    "generation_bytes_per_second": 468954.5379799801,
+    "upload_bytes_per_second": 4909153.902531052,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.31118149073431,
+    "estimated_full_shards_per_hour": 100.62672714757493,
     "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": 1.0372045380026975,
-      "bytes": 6042585,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.5961815709997609,
       "bytes": 6520796,
@@ -562,6 +556,12 @@
       "bytes": 12016198,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.4772214850017917,
+      "bytes": 1146549,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -385,3 +385,4 @@
 {"step": 683500, "loss": 0.49607446789741516, "lr": 0.0001, "grad_norm": 0.19096895211888984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100046773572, "tokens_per_second": 154159.41512214197, "data_wait_s": 1.0621995897963643e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 683600, "loss": 0.5936766862869263, "lr": 0.0001, "grad_norm": 0.21774702035682403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100061483139, "tokens_per_second": 153687.8288892015, "data_wait_s": 1.3689998013433069e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 683700, "loss": 0.5315948724746704, "lr": 0.0001, "grad_norm": 0.19818063535123745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100076217877, "tokens_per_second": 153717.3298400506, "data_wait_s": 1.282399898627773e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 683800, "loss": 0.5771541595458984, "lr": 0.0001, "grad_norm": 0.20099492334930003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100090848804, "tokens_per_second": 152418.9101668093, "data_wait_s": 1.1866002751048654e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:48862bc25ccbdaa0af71784ababbffb10c806f8f36d337c56714faea8275e8c2
-size 825312
+oid sha256:25f306953f14f23ba1ba238cb17945d44ec5c9d5f0a03834483ffd398e8c34fd
+size 827430