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": 76598.47085959032,
-      "required_pages_per_second_upper_estimate": 0.005299259947005673
+      "required_token_positions_per_second": 77720.15667700525,
+      "required_pages_per_second_upper_estimate": 0.005376860774524112
     },
     {
       "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": 9574.80885744879,
-      "required_pages_per_second_upper_estimate": 0.00400877252386839
+      "required_token_positions_per_second": 9715.019584625656,
+      "required_pages_per_second_upper_estimate": 0.004067475827404413
     },
     {
       "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": 19149.61771489758,
-      "required_pages_per_second_upper_estimate": 0.003988360360633614
+      "required_token_positions_per_second": 19430.03916925131,
+      "required_pages_per_second_upper_estimate": 0.0040467647543644385
     },
     {
       "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": 28724.42657234637,
-      "required_pages_per_second_upper_estimate": 0.0017219855133523455
+      "required_token_positions_per_second": 29145.05875387697,
+      "required_pages_per_second_upper_estimate": 0.0017472017703669524
     },
     {
       "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": 28724.42657234637,
-      "required_pages_per_second_upper_estimate": 0.002233197820600032
+      "required_token_positions_per_second": 29145.05875387697,
+      "required_pages_per_second_upper_estimate": 0.002265900122548601
     },
     {
       "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": 28724.42657234637,
-      "required_pages_per_second_upper_estimate": 0.36844610218373763
+      "required_token_positions_per_second": 29145.05875387697,
+      "required_pages_per_second_upper_estimate": 0.3738415201687635
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11153247267139932,
+    "producer_batch_seconds": 0.36095955026542015,
     "producer_p95_seconds": 0.3309544880030444,
-    "gpu_update_seconds": 0.9625257419974863,
-    "producer_batches_per_second": 8.965998655353346,
-    "gpu_batches_per_second": 1.0389332527613704,
-    "producer_headroom": 8.630004508492442,
+    "gpu_update_seconds": 0.9486342173292814,
+    "producer_batches_per_second": 2.770393522666686,
+    "gpu_batches_per_second": 1.0541470903456658,
+    "producer_headroom": 2.6280900910690224,
     "sustainable": true,
-    "buffer_seconds": 60.639121745841635,
+    "buffer_seconds": 60.71258990907401,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 446441.4582145016,
-    "upload_bytes_per_second": 4860803.681537214,
+    "generation_bytes_per_second": 406391.02295016363,
+    "upload_bytes_per_second": 4889098.734559173,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.79594430757795,
+    "estimated_full_shards_per_hour": 87.20205322626764,
     "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": 1.2482218100012688,
-      "bytes": 2427794,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.1882967479978106,
       "bytes": 6156483,
@@ -562,6 +556,12 @@
       "bytes": 1341787,
       "pages": 37,
       "shards": 1
+    },
+    {
+      "seconds": 1.4330705320026027,
+      "bytes": 5823183,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -393,3 +393,4 @@
 {"step": 684300, "loss": 0.5552197694778442, "lr": 0.0001, "grad_norm": 0.20512516525554436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100164277951, "tokens_per_second": 153968.53071679545, "data_wait_s": 9.00599843589589e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 684400, "loss": 0.5444086194038391, "lr": 0.0001, "grad_norm": 0.20616418955504132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100178984597, "tokens_per_second": 153473.657010538, "data_wait_s": 1.1593001545406878e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 684500, "loss": 0.5723017454147339, "lr": 0.0001, "grad_norm": 0.19126483496215455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100193504742, "tokens_per_second": 152234.8185075767, "data_wait_s": 1.79049966391176e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 684600, "loss": 0.497664213180542, "lr": 0.0001, "grad_norm": 0.21434360141424869, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100208048095, "tokens_per_second": 151214.19391769977, "data_wait_s": 1.1353004083503038e-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:cdcdce03f91ff16057e599666219952345f75f71cf9ec2ae078afaa1af990490
-size 842256
+oid sha256:1a7a1cc6a9ec0dedbed0869a290759fd481cc14248975ad7feee6e1e0f1c692d
+size 844374