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": 76907.0493337691,
-      "required_pages_per_second_upper_estimate": 0.005320608121850067
+      "required_token_positions_per_second": 76756.85820631939,
+      "required_pages_per_second_upper_estimate": 0.005310217551161667
     },
     {
       "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": 9613.381166721138,
-      "required_pages_per_second_upper_estimate": 0.00402492194427931
+      "required_token_positions_per_second": 9594.607275789924,
+      "required_pages_per_second_upper_estimate": 0.004017061708189834
     },
     {
       "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": 19226.762333442275,
-      "required_pages_per_second_upper_estimate": 0.004004427550236073
+      "required_token_positions_per_second": 19189.214551579847,
+      "required_pages_per_second_upper_estimate": 0.003996607337475725
     },
     {
       "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": 28840.143500163413,
-      "required_pages_per_second_upper_estimate": 0.0017289225664854592
+      "required_token_positions_per_second": 28783.82182736977,
+      "required_pages_per_second_upper_estimate": 0.0017255461681996952
     },
     {
       "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": 28840.143500163413,
-      "required_pages_per_second_upper_estimate": 0.0022421943027528327
+      "required_token_positions_per_second": 28783.82182736977,
+      "required_pages_per_second_upper_estimate": 0.0022378155404259845
     },
     {
       "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": 28840.143500163413,
-      "required_pages_per_second_upper_estimate": 0.3699303946866178
+      "required_token_positions_per_second": 28783.82182736977,
+      "required_pages_per_second_upper_estimate": 0.3692079607415217
     }
   ],
   "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": 0.16139050590624038,
-    "producer_p95_seconds": 0.3226698690050398,
-    "gpu_update_seconds": 0.9586637458944455,
-    "producer_batches_per_second": 6.196151343505601,
-    "gpu_batches_per_second": 1.043118616180679,
-    "producer_headroom": 5.9400256570939804,
+    "producer_batch_seconds": 0.09161349905508587,
+    "producer_p95_seconds": 0.32335810599761317,
+    "gpu_update_seconds": 0.960539575523298,
+    "producer_batches_per_second": 10.915421966349244,
+    "gpu_batches_per_second": 1.0410815186403997,
+    "producer_headroom": 10.484694782214786,
     "sustainable": true,
-    "buffer_seconds": 60.39581599135006,
+    "buffer_seconds": 60.513993257967776,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335251.3741439791,
-    "upload_bytes_per_second": 4135325.5279601137,
+    "generation_bytes_per_second": 357069.7174045004,
+    "upload_bytes_per_second": 4135803.3517437354,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.93714063872842,
+    "estimated_full_shards_per_hour": 76.6188491973997,
     "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": 1.4635059559986985,
-      "bytes": 7719999,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.2099752760004776,
       "bytes": 6423731,
@@ -562,6 +556,12 @@
       "bytes": 1071557,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.4824117819953244,
+      "bytes": 7847757,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -336,3 +336,4 @@
 {"step": 876100, "loss": 0.5254486203193665, "lr": 3e-05, "grad_norm": 0.30969942609753287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128236336868, "tokens_per_second": 151478.03414473226, "data_wait_s": 1.1274001735728234e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 876200, "loss": 0.5402370691299438, "lr": 3e-05, "grad_norm": 0.3122375962551669, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128250930986, "tokens_per_second": 151039.32049315082, "data_wait_s": 1.5830002666916698e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 876300, "loss": 0.5127866268157959, "lr": 3e-05, "grad_norm": 0.3174092561154268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128265649775, "tokens_per_second": 152528.61273359173, "data_wait_s": 9.88500687526539e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 876400, "loss": 0.5109633207321167, "lr": 3e-05, "grad_norm": 0.3099059393556759, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128280329031, "tokens_per_second": 152866.22827241596, "data_wait_s": 1.052000152412802e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e06ae08aa3b06e2ae152dd2d47c7e6503ae865eaa24177eea89a42820beb50f2
-size 721530
+oid sha256:a11154e62b9107b4ac9b04ca51db5f8c5cefdba908187eff38b16ea22a4f00bc
+size 723648