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": 76995.49654822006,
-      "required_pages_per_second_upper_estimate": 0.005326727105371597
+      "required_token_positions_per_second": 76850.35198907083,
+      "required_pages_per_second_upper_estimate": 0.005316685667987876
     },
     {
       "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": 9624.437068527508,
-      "required_pages_per_second_upper_estimate": 0.004029550819492154
+      "required_token_positions_per_second": 9606.293998633853,
+      "required_pages_per_second_upper_estimate": 0.004021954694998063
     },
     {
       "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": 19248.874137055016,
-      "required_pages_per_second_upper_estimate": 0.004009032855801132
+      "required_token_positions_per_second": 19212.587997267707,
+      "required_pages_per_second_upper_estimate": 0.004001475409813293
     },
     {
       "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": 28873.311205582526,
-      "required_pages_per_second_upper_estimate": 0.0017309109197811811
+      "required_token_positions_per_second": 28818.88199590156,
+      "required_pages_per_second_upper_estimate": 0.0017276479717693984
     },
     {
       "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": 28873.311205582526,
-      "required_pages_per_second_upper_estimate": 0.0022447729459598507
+      "required_token_positions_per_second": 28818.88199590156,
+      "required_pages_per_second_upper_estimate": 0.0022405413143159474
     },
     {
       "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": 28873.311205582526,
-      "required_pages_per_second_upper_estimate": 0.37035583439902675
+      "required_token_positions_per_second": 28818.88199590156,
+      "required_pages_per_second_upper_estimate": 0.36965767493877144
     }
   ],
   "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": 0.2324724740153954,
-    "producer_p95_seconds": 0.2857396469989908,
-    "gpu_update_seconds": 0.9575624978770839,
-    "producer_batches_per_second": 4.301584539138924,
-    "gpu_batches_per_second": 1.044318258303766,
-    "producer_headroom": 4.119036036127313,
+    "producer_batch_seconds": 0.07497230774987429,
+    "producer_p95_seconds": 0.2577247459994396,
+    "gpu_update_seconds": 0.9593710125164441,
+    "producer_batches_per_second": 13.338258218437684,
+    "gpu_batches_per_second": 1.0423496092267637,
+    "producer_headroom": 12.796338292228342,
     "sustainable": true,
-    "buffer_seconds": 60.326437366256286,
+    "buffer_seconds": 60.44037378853598,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441038.6304884399,
-    "upload_bytes_per_second": 4642072.979772448,
+    "generation_bytes_per_second": 443335.35009692895,
+    "upload_bytes_per_second": 4765505.842011201,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.63662324895762,
+    "estimated_full_shards_per_hour": 95.1294458120432,
     "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": 0.8775070329993468,
-      "bytes": 1128371,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.7969425399987813,
       "bytes": 12329538,
@@ -562,6 +556,12 @@
       "bytes": 2400719,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.58055241799957,
+      "bytes": 15825794,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -235,3 +235,4 @@
 {"step": 463700, "loss": 0.5837123394012451, "lr": 0.0003, "grad_norm": 0.15454984864182733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67871826030, "tokens_per_second": 151179.7593776281, "data_wait_s": 1.3089000276522711e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 463800, "loss": 0.598275899887085, "lr": 0.0003, "grad_norm": 0.13880101897746747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67886414821, "tokens_per_second": 151588.37576327668, "data_wait_s": 8.752998837735504e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 463900, "loss": 0.5890750288963318, "lr": 0.0003, "grad_norm": 0.1425417499480492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67900955424, "tokens_per_second": 150741.81136325275, "data_wait_s": 1.1448002624092624e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 464000, "loss": 0.5789682269096375, "lr": 0.0003, "grad_norm": 0.15111280592420623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67915593951, "tokens_per_second": 150678.78471906632, "data_wait_s": 1.577700095367618e-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 sha256:9ed04e6e86f9aedecddd2694ad0f16ccf8e161f1ace7cdfdb242ccf0dbdef542
-size 505263
+oid sha256:c401c4779212ec35c21322219485af74a78cdf8137b4f3cd52a2922a2fc2873d
+size 507381