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": 77230.82575529839,
-      "required_pages_per_second_upper_estimate": 0.005343007725956265
+      "required_token_positions_per_second": 77382.8199354448,
+      "required_pages_per_second_upper_estimate": 0.005353523036013881
     },
     {
       "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": 9653.853219412298,
-      "required_pages_per_second_upper_estimate": 0.0040418667475885335
+      "required_token_positions_per_second": 9672.8524919306,
+      "required_pages_per_second_upper_estimate": 0.00404982134624202
     },
     {
       "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": 19307.706438824596,
-      "required_pages_per_second_upper_estimate": 0.004021286072747603
+      "required_token_positions_per_second": 19345.7049838612,
+      "required_pages_per_second_upper_estimate": 0.004029200167589708
     },
     {
       "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": 28961.559658236893,
-      "required_pages_per_second_upper_estimate": 0.0017362012797702254
+      "required_token_positions_per_second": 29018.557475791797,
+      "required_pages_per_second_upper_estimate": 0.0017396182119020138
     },
     {
       "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": 28961.559658236893,
-      "required_pages_per_second_upper_estimate": 0.002251633874989807
+      "required_token_positions_per_second": 29018.557475791797,
+      "required_pages_per_second_upper_estimate": 0.002256065204597797
     },
     {
       "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": 28961.559658236893,
-      "required_pages_per_second_upper_estimate": 0.3714877907958709
+      "required_token_positions_per_second": 29018.557475791797,
+      "required_pages_per_second_upper_estimate": 0.3722188975999769
     }
   ],
   "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": 0.08078131803895872,
-    "producer_p95_seconds": 0.20612914499906765,
-    "gpu_update_seconds": 0.9546447196305152,
-    "producer_batches_per_second": 12.379099825008131,
-    "gpu_batches_per_second": 1.0475101149535913,
-    "producer_headroom": 11.817642281723048,
+    "producer_batch_seconds": 0.2996023386092759,
+    "producer_p95_seconds": 0.20595333600067534,
+    "gpu_update_seconds": 0.9527696207182194,
+    "producer_batches_per_second": 3.3377576578403896,
+    "gpu_batches_per_second": 1.0495716679612195,
+    "producer_headroom": 3.1801140977099203,
     "sustainable": true,
-    "buffer_seconds": 60.14261733672246,
+    "buffer_seconds": 60.02448610524782,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 527213.3654222961,
-    "upload_bytes_per_second": 5814995.354598551,
+    "generation_bytes_per_second": 469344.59499972244,
+    "upload_bytes_per_second": 5808767.891247333,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.12771532060302,
+    "estimated_full_shards_per_hour": 100.71042430394893,
     "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": 1.0907007900004828,
-      "bytes": 6092133,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8937167260000933,
       "bytes": 2214439,
@@ -562,6 +556,12 @@
       "bytes": 6068911,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1270621679996111,
+      "bytes": 5842224,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -133,3 +133,4 @@
 {"step": 538300, "loss": 0.5773850083351135, "lr": 0.0003, "grad_norm": 0.15849421215983392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78789289289, "tokens_per_second": 150842.4074431509, "data_wait_s": 1.0429999747429974e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 538400, "loss": 0.5680321455001831, "lr": 0.0003, "grad_norm": 0.1443809803251855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78803872259, "tokens_per_second": 152702.5826904807, "data_wait_s": 1.6085999959614128e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 538500, "loss": 0.5491064786911011, "lr": 0.0003, "grad_norm": 0.15972445504468055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78818488987, "tokens_per_second": 153061.73812530036, "data_wait_s": 1.0843999916687608e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 538600, "loss": 0.5387080311775208, "lr": 0.0003, "grad_norm": 0.16702672484838227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78833117219, "tokens_per_second": 152753.8111222553, "data_wait_s": 1.3556998965214007e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bac1c6c91a9cd3f91c601a96524fc57cbf102eb58ef6860bb4e76318f5b5a784
-size 287661
+oid sha256:07759224d42a1f37e73eb9cd9631d1a0ad6034efd3af0bc2d29f1c5fc62d93a5
+size 289779