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": 76677.57584239969,
-      "required_pages_per_second_upper_estimate": 0.005304732613265273
+      "required_token_positions_per_second": 77077.9503837055,
+      "required_pages_per_second_upper_estimate": 0.005332431453029736
     },
     {
       "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": 9584.696980299961,
-      "required_pages_per_second_upper_estimate": 0.00401291247441869
+      "required_token_positions_per_second": 9634.743797963187,
+      "required_pages_per_second_upper_estimate": 0.004033866031877887
     },
     {
       "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": 19169.393960599922,
-      "required_pages_per_second_upper_estimate": 0.003992479231078741
+      "required_token_positions_per_second": 19269.487595926374,
+      "required_pages_per_second_upper_estimate": 0.004013326095670569
     },
     {
       "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": 28754.09094089988,
-      "required_pages_per_second_upper_estimate": 0.0017237638469522604
+      "required_token_positions_per_second": 28904.23139388956,
+      "required_pages_per_second_upper_estimate": 0.001732764537857794
     },
     {
       "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": 28754.09094089988,
-      "required_pages_per_second_upper_estimate": 0.0022355040947752997
+      "required_token_positions_per_second": 28904.23139388956,
+      "required_pages_per_second_upper_estimate": 0.0022471768545971878
     },
     {
       "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": 28754.09094089988,
-      "required_pages_per_second_upper_estimate": 0.3688266048524247
+      "required_token_positions_per_second": 28904.23139388956,
+      "required_pages_per_second_upper_estimate": 0.3707524453751178
     }
   ],
   "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": 0.0884925625625641,
-    "producer_p95_seconds": 0.20395748000009917,
-    "gpu_update_seconds": 0.9615327452648981,
-    "producer_batches_per_second": 11.30038469948253,
-    "gpu_batches_per_second": 1.0400061827582423,
-    "producer_headroom": 10.865689922642888,
+    "producer_batch_seconds": 0.07666722103132884,
+    "producer_p95_seconds": 0.21711217899974145,
+    "gpu_update_seconds": 0.9565381491460405,
+    "producer_batches_per_second": 13.043383946202589,
+    "gpu_batches_per_second": 1.0454366100220471,
+    "producer_headroom": 12.476494338501801,
     "sustainable": true,
-    "buffer_seconds": 60.57656295168858,
+    "buffer_seconds": 60.26190339620055,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 446179.9099657071,
-    "upload_bytes_per_second": 5649463.947394049,
+    "generation_bytes_per_second": 464095.759317478,
+    "upload_bytes_per_second": 5591223.821528959,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.73982214191828,
+    "estimated_full_shards_per_hour": 99.58414635318046,
     "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": 1.185791879000135,
-      "bytes": 7588973,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8400346700000227,
       "bytes": 806070,
@@ -562,6 +556,12 @@
       "bytes": 7559192,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.825367109000581,
+      "bytes": 1141341,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -143,3 +143,4 @@
 {"step": 774300, "loss": 0.5189425945281982, "lr": 3e-05, "grad_norm": 0.25102008123817554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113335282156, "tokens_per_second": 152687.81659912478, "data_wait_s": 2.1437999748741277e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 774400, "loss": 0.5151251554489136, "lr": 3e-05, "grad_norm": 0.2618655086145514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113349957380, "tokens_per_second": 152664.69233507168, "data_wait_s": 1.537999924039468e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 774500, "loss": 0.5222188234329224, "lr": 3e-05, "grad_norm": 0.24253370722559092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113364531091, "tokens_per_second": 151652.01044864577, "data_wait_s": 1.0929999916697852e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 774600, "loss": 0.5660772323608398, "lr": 3e-05, "grad_norm": 0.2558079389723242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113379147847, "tokens_per_second": 151113.6816932582, "data_wait_s": 9.150999176199548e-06, "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 sha256:589f428ed3ecfc20726888f3886c9ea687eae81111ce21473617b4cba8f70bb7
-size 308841
+oid sha256:540c96a270e3d9104778addf7a6f47a982768cd86db5c49f09f2515ad5e8bd16
+size 310959