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": 76555.0266324517,
-      "required_pages_per_second_upper_estimate": 0.005296254374567728
+      "required_token_positions_per_second": 77086.56362944118,
+      "required_pages_per_second_upper_estimate": 0.005333027337355214
     },
     {
       "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": 9569.378329056462,
-      "required_pages_per_second_upper_estimate": 0.004006498875032851
+      "required_token_positions_per_second": 9635.820453680148,
+      "required_pages_per_second_upper_estimate": 0.004034316805143451
     },
     {
       "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": 19138.756658112925,
-      "required_pages_per_second_upper_estimate": 0.003986098288930699
+      "required_token_positions_per_second": 19271.640907360295,
+      "required_pages_per_second_upper_estimate": 0.004013774573655614
     },
     {
       "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": 28708.134987169386,
-      "required_pages_per_second_upper_estimate": 0.0017210088576967955
+      "required_token_positions_per_second": 28907.461361040445,
+      "required_pages_per_second_upper_estimate": 0.0017329581694566153
     },
     {
       "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": 28708.134987169386,
-      "required_pages_per_second_upper_estimate": 0.0022319312215116315
+      "required_token_positions_per_second": 28907.461361040445,
+      "required_pages_per_second_upper_estimate": 0.0022474279703360448
     },
     {
       "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": 28708.134987169386,
-      "required_pages_per_second_upper_estimate": 0.3682371312216286
+      "required_token_positions_per_second": 28907.461361040445,
+      "required_pages_per_second_upper_estimate": 0.3707938759256608
     }
   ],
   "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": 0.8933466412108544,
-    "producer_p95_seconds": 0.2676944689992524,
-    "gpu_update_seconds": 0.9630719659203499,
-    "producer_batches_per_second": 1.1193863097135353,
-    "gpu_batches_per_second": 1.0383440027188,
-    "producer_headroom": 1.0780495739201401,
+    "producer_batch_seconds": 0.08031042316440562,
+    "producer_p95_seconds": 0.2767030530012562,
+    "gpu_update_seconds": 0.9564312706221286,
+    "producer_batches_per_second": 12.451683861170464,
+    "gpu_batches_per_second": 1.0455534346441133,
+    "producer_headroom": 11.909179816724318,
     "sustainable": true,
-    "buffer_seconds": 60.67353385298205,
+    "buffer_seconds": 60.255170049194106,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 235841.00489855447,
-    "upload_bytes_per_second": 4597870.363164352,
+    "generation_bytes_per_second": 234700.51200661343,
+    "upload_bytes_per_second": 4719289.395295465,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.6059895536182,
+    "estimated_full_shards_per_hour": 50.36126632832338,
     "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": 1.584322863000125,
-      "bytes": 6087710,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2062028339969402,
       "bytes": 5756694,
@@ -562,6 +556,12 @@
       "bytes": 2149233,
       "pages": 21,
       "shards": 2
+    },
+    {
+      "seconds": 1.4335154450018308,
+      "bytes": 16606396,
+      "pages": 45,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -292,3 +292,4 @@
 {"step": 914200, "loss": 0.5317642092704773, "lr": 3e-05, "grad_norm": 0.3284383063020626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133812466888, "tokens_per_second": 151018.95310672675, "data_wait_s": 9.747000149218366e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 914300, "loss": 0.511695384979248, "lr": 3e-05, "grad_norm": 0.3179288216994619, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133826971546, "tokens_per_second": 123452.86378734301, "data_wait_s": 1.456199970562011e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 914400, "loss": 0.5224106907844543, "lr": 3e-05, "grad_norm": 0.35033095834050637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133841463019, "tokens_per_second": 151270.9553593144, "data_wait_s": 1.5842000721022487e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 914500, "loss": 0.5214608907699585, "lr": 3e-05, "grad_norm": 0.3191385382969053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133856084120, "tokens_per_second": 151598.96451312004, "data_wait_s": 1.6672001947881654e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d98d7229e3a2f50b6ab69a85dada5d82f1827e155fbfd82a6e54f4e94a201eeb
-size 626772
+oid sha256:6409506d437c6ad4564235fc3b84f8b5da06292266b59f4826a9a58028afd072
+size 628890