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": 76666.63139458954,
-      "required_pages_per_second_upper_estimate": 0.005303975450971149
+      "required_token_positions_per_second": 76598.0376333891,
+      "required_pages_per_second_upper_estimate": 0.005299229975411849
     },
     {
       "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": 9583.328924323692,
-      "required_pages_per_second_upper_estimate": 0.0040123396979496835
+      "required_token_positions_per_second": 9574.754704173638,
+      "required_pages_per_second_upper_estimate": 0.0040087498510229325
     },
     {
       "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": 19166.657848647385,
-      "required_pages_per_second_upper_estimate": 0.003991909371115152
+      "required_token_positions_per_second": 19149.509408347276,
+      "required_pages_per_second_upper_estimate": 0.0039883378032354215
     },
     {
       "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": 28749.98677297108,
-      "required_pages_per_second_upper_estimate": 0.0017235178083516304
+      "required_token_positions_per_second": 28724.264112520916,
+      "required_pages_per_second_upper_estimate": 0.0017219757741338713
     },
     {
       "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": 28749.98677297108,
-      "required_pages_per_second_upper_estimate": 0.0022351850137711624
+      "required_token_positions_per_second": 28724.264112520916,
+      "required_pages_per_second_upper_estimate": 0.0022331851900631882
     },
     {
       "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": 28749.98677297108,
-      "required_pages_per_second_upper_estimate": 0.3687739609929462
+      "required_token_positions_per_second": 28724.264112520916,
+      "required_pages_per_second_upper_estimate": 0.36844401832353246
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09465892570341339,
-    "producer_p95_seconds": 0.3426627410008223,
-    "gpu_update_seconds": 0.961670007653461,
-    "producer_batches_per_second": 10.56424412773512,
-    "gpu_batches_per_second": 1.0398577391844286,
-    "producer_headroom": 10.159316731172064,
+    "producer_batch_seconds": 0.11446951010952944,
+    "producer_p95_seconds": 0.3758285910007544,
+    "gpu_update_seconds": 0.9625311858885265,
+    "producer_batches_per_second": 8.735950726469923,
+    "gpu_batches_per_second": 1.038927376754952,
+    "producer_headroom": 8.40862501261283,
     "sustainable": true,
-    "buffer_seconds": 60.585210482168044,
+    "buffer_seconds": 60.639464710977165,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380783.52992785745,
-    "upload_bytes_per_second": 5033920.350783255,
+    "generation_bytes_per_second": 480860.6981868346,
+    "upload_bytes_per_second": 5006361.88017286,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.70728133560937,
+    "estimated_full_shards_per_hour": 103.18151196674135,
     "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": 1.2491005060001044,
-      "bytes": 11541986,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.6100285119973705,
       "bytes": 7404435,
@@ -562,6 +556,12 @@
       "bytes": 2440099,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.5953800970019074,
+      "bytes": 10985431,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -365,3 +365,4 @@
 {"step": 476700, "loss": 0.6041773557662964, "lr": 0.0003, "grad_norm": 0.18339025125092456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69773730765, "tokens_per_second": 151307.86727909005, "data_wait_s": 1.4555997040588409e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 476800, "loss": 0.6477830410003662, "lr": 0.0003, "grad_norm": 0.17604146039098187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69788381008, "tokens_per_second": 151402.06029228034, "data_wait_s": 9.354000212624669e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 476900, "loss": 0.6106101274490356, "lr": 0.0003, "grad_norm": 0.16665720833439113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69802927842, "tokens_per_second": 150698.06712113414, "data_wait_s": 9.061004675459117e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 477000, "loss": 0.6186416745185852, "lr": 0.0003, "grad_norm": 0.1934919045075562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69817582261, "tokens_per_second": 151649.48824863945, "data_wait_s": 1.2336000509094447e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e4dea0d322b27012ab0b3e3b23773524d13d5c57ed8dbde9e2f11fbbf7ade3b
-size 782952
+oid sha256:bbb346dfbbef00df1b07d9cabce2e49164d912322fc013db368ac26f3503807c
+size 785070