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": 77264.92382139551,
-      "required_pages_per_second_upper_estimate": 0.005345366709287277
+      "required_token_positions_per_second": 76950.11810001342,
+      "required_pages_per_second_upper_estimate": 0.005323587719032154
     },
     {
       "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": 9658.115477674439,
-      "required_pages_per_second_upper_estimate": 0.0040436512661168175
+      "required_token_positions_per_second": 9618.764762501678,
+      "required_pages_per_second_upper_estimate": 0.0040271759434103585
     },
     {
       "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": 19316.230955348878,
-      "required_pages_per_second_upper_estimate": 0.004023061504732993
+      "required_token_positions_per_second": 19237.529525003356,
+      "required_pages_per_second_upper_estimate": 0.00400667007228831
     },
     {
       "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": 28974.346433023315,
-      "required_pages_per_second_upper_estimate": 0.0017369678273943968
+      "required_token_positions_per_second": 28856.294287505036,
+      "required_pages_per_second_upper_estimate": 0.001729890781525765
     },
     {
       "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": 28974.346433023315,
-      "required_pages_per_second_upper_estimate": 0.002252627990486373
+      "required_token_positions_per_second": 28856.294287505036,
+      "required_pages_per_second_upper_estimate": 0.0022434499554288383
     },
     {
       "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": 28974.346433023315,
-      "required_pages_per_second_upper_estimate": 0.371651805813462
+      "required_token_positions_per_second": 28856.294287505036,
+      "required_pages_per_second_upper_estimate": 0.3701375596452718
     }
   ],
   "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": 0.10970470340623706,
-    "producer_p95_seconds": 0.2854320159967756,
-    "gpu_update_seconds": 0.9542234218780645,
-    "producer_batches_per_second": 9.115379459137637,
-    "gpu_batches_per_second": 1.0479725995740492,
-    "producer_headroom": 8.698108579215337,
+    "producer_batch_seconds": 0.6660843871951556,
+    "producer_p95_seconds": 0.28682142899924656,
+    "gpu_update_seconds": 0.958127184472601,
+    "producer_batches_per_second": 1.5013112741028873,
+    "gpu_batches_per_second": 1.0437027737089495,
+    "producer_headroom": 1.4384471440731728,
     "sustainable": true,
-    "buffer_seconds": 60.11607557831807,
+    "buffer_seconds": 60.362012621773864,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362916.5337134667,
-    "upload_bytes_per_second": 4756601.556279353,
+    "generation_bytes_per_second": 313577.9340428975,
+    "upload_bytes_per_second": 4878243.809200962,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.87343987038614,
+    "estimated_full_shards_per_hour": 67.2865249249,
     "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": 2.1865444229988498,
-      "bytes": 7544891,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.1021929470007308,
       "bytes": 6736997,
@@ -562,6 +556,12 @@
       "bytes": 2775913,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.592129067001224,
+      "bytes": 15451622,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -302,3 +302,4 @@
 {"step": 790200, "loss": 0.5087929964065552, "lr": 3e-05, "grad_norm": 0.2584321778377075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115662904098, "tokens_per_second": 153086.08349205996, "data_wait_s": 9.760999091668054e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 790300, "loss": 0.5574849247932434, "lr": 3e-05, "grad_norm": 0.28539266337026326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115677722134, "tokens_per_second": 154485.94780169296, "data_wait_s": 8.484999852953479e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 790400, "loss": 0.4686794579029083, "lr": 3e-05, "grad_norm": 0.2493854993032542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115692214851, "tokens_per_second": 150570.18944420552, "data_wait_s": 1.400900146109052e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 790500, "loss": 0.5351627469062805, "lr": 3e-05, "grad_norm": 0.26463694846194785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115706737281, "tokens_per_second": 149942.46834166712, "data_wait_s": 1.6002002666937187e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2bb8e5c152bb5bac7f13f85ea77905a99c16fb007dd2401a21b4c1886174182e
-size 648735
+oid sha256:8823089a87822c47846b240d96a315746b6fe8fb2f44d444884d90265a809e17
+size 650853