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": 76581.33219895158,
-      "required_pages_per_second_upper_estimate": 0.005298074254695514
+      "required_token_positions_per_second": 76246.92987731703,
+      "required_pages_per_second_upper_estimate": 0.005274939526164548
     },
     {
       "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": 9572.666524868948,
-      "required_pages_per_second_upper_estimate": 0.004007875573954201
+      "required_token_positions_per_second": 9530.866234664629,
+      "required_pages_per_second_upper_estimate": 0.003990374665334972
     },
     {
       "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": 19145.333049737896,
-      "required_pages_per_second_upper_estimate": 0.003987467977875071
+      "required_token_positions_per_second": 19061.732469329258,
+      "required_pages_per_second_upper_estimate": 0.003970056181671516
     },
     {
       "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": 28717.999574606845,
-      "required_pages_per_second_upper_estimate": 0.0017216002246513184
+      "required_token_positions_per_second": 28592.59870399389,
+      "required_pages_per_second_upper_estimate": 0.0017140826339341142
     },
     {
       "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": 28717.999574606845,
-      "required_pages_per_second_upper_estimate": 0.0022326981497951593
+      "required_token_positions_per_second": 28592.59870399389,
+      "required_pages_per_second_upper_estimate": 0.002222948783685139
     },
     {
       "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": 28717.999574606845,
-      "required_pages_per_second_upper_estimate": 0.36836366355750755
+      "required_token_positions_per_second": 28592.59870399389,
+      "required_pages_per_second_upper_estimate": 0.366755155834249
     }
   ],
   "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": 0.107902268164068,
-    "producer_p95_seconds": 0.23770479000086198,
-    "gpu_update_seconds": 0.9627411522231177,
-    "producer_batches_per_second": 9.267645778117249,
-    "gpu_batches_per_second": 1.038700794799148,
-    "producer_headroom": 8.922343974820311,
+    "producer_batch_seconds": 0.11856429832025128,
+    "producer_p95_seconds": 0.21887353300007817,
+    "gpu_update_seconds": 0.9669635238904695,
+    "producer_batches_per_second": 8.434242129944742,
+    "gpu_batches_per_second": 1.0341651730321864,
+    "producer_headroom": 8.155604491316828,
     "sustainable": true,
-    "buffer_seconds": 60.65269259005642,
+    "buffer_seconds": 60.918702005099576,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344835.3733125481,
-    "upload_bytes_per_second": 4000830.3600562094,
+    "generation_bytes_per_second": 395771.85115218355,
+    "upload_bytes_per_second": 4032813.9967123894,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.99364375622112,
+    "estimated_full_shards_per_hour": 84.92342616008882,
     "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": 1.2644712469991646,
-      "bytes": 6316580,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.6103071739998995,
       "bytes": 7727369,
@@ -562,6 +556,12 @@
       "bytes": 1225243,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.7503577689985832,
+      "bytes": 11722749,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -122,3 +122,4 @@
 {"step": 812200, "loss": 0.5729753375053406, "lr": 3e-05, "grad_norm": 0.29566897251418217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118882688269, "tokens_per_second": 153201.9355423492, "data_wait_s": 1.7569000192452222e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 812300, "loss": 0.5137185454368591, "lr": 3e-05, "grad_norm": 0.2792861971277968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118897314241, "tokens_per_second": 151032.37952066955, "data_wait_s": 1.7754000509739853e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 812400, "loss": 0.5304609537124634, "lr": 3e-05, "grad_norm": 0.2786341261923443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118912177957, "tokens_per_second": 154063.14123069774, "data_wait_s": 1.5179999536485411e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 812500, "loss": 0.53594571352005, "lr": 3e-05, "grad_norm": 0.2830846785903361, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118926928546, "tokens_per_second": 151748.15153972426, "data_wait_s": 1.848800093284808e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cca322e913d189c0950044d09a29c3063d85a774aec3058fb7304e60936108c1
-size 264363
+oid sha256:64b348bb55cdedab1c9d6a20a0ac9322ac3d9563500fd033a1230ee6a29bdf6d
+size 266481