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": 77444.12768165825,
-      "required_pages_per_second_upper_estimate": 0.005357764448150498
+      "required_token_positions_per_second": 77161.90788750623,
+      "required_pages_per_second_upper_estimate": 0.005338239827950919
     },
     {
       "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": 9680.515960207282,
-      "required_pages_per_second_upper_estimate": 0.004053029880378051
+      "required_token_positions_per_second": 9645.238485938278,
+      "required_pages_per_second_upper_estimate": 0.004038259938579053
     },
     {
       "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": 19361.031920414564,
-      "required_pages_per_second_upper_estimate": 0.004032392364275275
+      "required_token_positions_per_second": 19290.476971876556,
+      "required_pages_per_second_upper_estimate": 0.004017697629153925
     },
     {
       "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": 29041.547880621845,
-      "required_pages_per_second_upper_estimate": 0.0017409964515672596
+      "required_token_positions_per_second": 28935.715457814833,
+      "required_pages_per_second_upper_estimate": 0.001734651959416732
     },
     {
       "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": 29041.547880621845,
-      "required_pages_per_second_upper_estimate": 0.0022578526074492292
+      "required_token_positions_per_second": 28935.715457814833,
+      "required_pages_per_second_upper_estimate": 0.0022496246020836156
     },
     {
       "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": 29041.547880621845,
-      "required_pages_per_second_upper_estimate": 0.3725137938279633
+      "required_token_positions_per_second": 28935.715457814833,
+      "required_pages_per_second_upper_estimate": 0.37115628914219717
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07481326718749415,
-    "producer_p95_seconds": 0.18776458600041224,
-    "gpu_update_seconds": 0.9520153716892032,
-    "producer_batches_per_second": 13.366613136862988,
-    "gpu_batches_per_second": 1.0504032074877694,
-    "producer_headroom": 12.725221173716404,
+    "producer_batch_seconds": 0.08787082855459971,
+    "producer_p95_seconds": 0.19416214999910153,
+    "gpu_update_seconds": 0.955497369343012,
+    "producer_batches_per_second": 11.380341080756246,
+    "gpu_batches_per_second": 1.046575356547122,
+    "producer_headroom": 10.873885964888801,
     "sustainable": true,
-    "buffer_seconds": 60.92898378810901,
+    "buffer_seconds": 60.19633426860975,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276923.3195714457,
-    "upload_bytes_per_second": 3700582.7276740214,
+    "generation_bytes_per_second": 267483.30610315915,
+    "upload_bytes_per_second": 3719851.275735463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.421297935080794,
+    "estimated_full_shards_per_hour": 57.395690797053156,
     "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": 1.2620071970000026,
-      "bytes": 1106661,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 33.80818266799997,
       "bytes": 13381894,
@@ -562,6 +556,12 @@
       "bytes": 11002922,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 0.9981318899990583,
+      "bytes": 2367436,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -111,3 +111,4 @@
 {"step": 656100, "loss": 0.537526547908783, "lr": 0.0001, "grad_norm": 0.17528571990657263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96035457912, "tokens_per_second": 150510.09848665766, "data_wait_s": 1.6138999853865243e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 656200, "loss": 0.5577178597450256, "lr": 0.0001, "grad_norm": 0.17811262485610355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96050104145, "tokens_per_second": 153298.7697921858, "data_wait_s": 1.5414001609315164e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 656300, "loss": 0.5355839729309082, "lr": 0.0001, "grad_norm": 0.18347786089207452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96064810224, "tokens_per_second": 153915.89342155104, "data_wait_s": 1.596100082679186e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 656400, "loss": 0.5306210517883301, "lr": 0.0001, "grad_norm": 0.19070432796215145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96079568729, "tokens_per_second": 154493.6201685895, "data_wait_s": 1.0431000191601925e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4a775157317a2c2b085307cc08ee9b3aa7cf4ee57d920a6353a4a5428b1b01be
-size 241065
+oid sha256:77dc768459d477fb9325fc50798bf222d24a33d0032d3609c233bb50abb7a88e
+size 243183