Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75748.84188281285,
-      "required_pages_per_second_upper_estimate": 0.006565717630717666
+      "required_token_positions_per_second": 75614.28557871691,
+      "required_pages_per_second_upper_estimate": 0.006554054631308465
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9468.605235351606,
-      "required_pages_per_second_upper_estimate": 0.003777168196645766
+      "required_token_positions_per_second": 9451.785697339614,
+      "required_pages_per_second_upper_estimate": 0.003770458631458279
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18937.210470703212,
-      "required_pages_per_second_upper_estimate": 0.00438079035887595
+      "required_token_positions_per_second": 18903.571394679228,
+      "required_pages_per_second_upper_estimate": 0.004373008550665318
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28405.81570605482,
-      "required_pages_per_second_upper_estimate": 0.0017297651427016674
+      "required_token_positions_per_second": 28355.357092018843,
+      "required_pages_per_second_upper_estimate": 0.0017266924778427646
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28405.81570605482,
-      "required_pages_per_second_upper_estimate": 0.0021867583253826674
+      "required_token_positions_per_second": 28355.357092018843,
+      "required_pages_per_second_upper_estimate": 0.0021828738815957943
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28405.81570605482,
-      "required_pages_per_second_upper_estimate": 0.3527051629195875
+      "required_token_positions_per_second": 28355.357092018843,
+      "required_pages_per_second_upper_estimate": 0.3520786358073785
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.15662519832032373,
-    "producer_p95_seconds": 0.16940903000022445,
-    "gpu_update_seconds": 0.9733218114946339,
-    "producer_batches_per_second": 6.384668691399446,
-    "gpu_batches_per_second": 1.0274094222386725,
-    "producer_headroom": 6.214337296505982,
+    "producer_batch_seconds": 0.09507295642186264,
+    "producer_p95_seconds": 0.16969986499998413,
+    "gpu_update_seconds": 0.9750538464487213,
+    "producer_batches_per_second": 10.518238178717702,
+    "gpu_batches_per_second": 1.0255843855620241,
+    "producer_headroom": 10.255848594022488,
     "sustainable": true,
-    "buffer_seconds": 60.3459523126673,
+    "buffer_seconds": 60.45333847982072,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441801.8864497533,
-    "upload_bytes_per_second": 3627618.272815116,
+    "generation_bytes_per_second": 432654.5516061851,
+    "upload_bytes_per_second": 3601035.366902879,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.80040021056604,
+    "estimated_full_shards_per_hour": 92.83759509219327,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 7164962,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0547767480002221,
+      "bytes": 2359076,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -30,3 +30,4 @@
 {"step": 3000, "loss": 0.8225421886891127, "lr": 0.0003, "grad_norm": 0.20396286009016903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 439054516, "tokens_per_second": 148361.7991957803, "data_wait_s": 1.125900007536984e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 3100, "loss": 0.8211501948535442, "lr": 0.0003, "grad_norm": 0.21141356178096296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 453545230, "tokens_per_second": 147444.06545955865, "data_wait_s": 1.2337000043771695e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 3200, "loss": 0.8346284031867981, "lr": 0.0003, "grad_norm": 0.20888028264026168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 468054322, "tokens_per_second": 148392.07020325723, "data_wait_s": 1.3474999832396861e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 3300, "loss": 0.8051414489746094, "lr": 0.0003, "grad_norm": 0.25193659655952183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 482570487, "tokens_per_second": 147370.6245055717, "data_wait_s": 9.698000212665647e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:edc6082b7803ae22c1ebcd6d59b6a8568cbc0e1a4b416c8177402779bacfa8de
-size 66782
+oid sha256:78d3a131938607cd93a4f0c437ff4911bc91caa19fdedeaeea7d96e261b0866c
+size 68865