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": 76577.43011913888,
-      "required_pages_per_second_upper_estimate": 0.005297804299759999
+      "required_token_positions_per_second": 77499.13812160284,
+      "required_pages_per_second_upper_estimate": 0.00536157019802767
     },
     {
       "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": 9572.17876489236,
-      "required_pages_per_second_upper_estimate": 0.00400767135903759
+      "required_token_positions_per_second": 9687.392265200355,
+      "required_pages_per_second_upper_estimate": 0.0040559088456850745
     },
     {
       "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": 19144.35752978472,
-      "required_pages_per_second_upper_estimate": 0.003987264802795015
+      "required_token_positions_per_second": 19374.78453040071,
+      "required_pages_per_second_upper_estimate": 0.004035256670254675
     },
     {
       "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": 28716.536294677077,
-      "required_pages_per_second_upper_estimate": 0.0017215125032538278
+      "required_token_positions_per_second": 29062.176795601066,
+      "required_pages_per_second_upper_estimate": 0.0017422331235217347
     },
     {
       "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": 28716.536294677077,
-      "required_pages_per_second_upper_estimate": 0.002232584386216908
+      "required_token_positions_per_second": 29062.176795601066,
+      "required_pages_per_second_upper_estimate": 0.0022594564148518564
     },
     {
       "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": 28716.536294677077,
-      "required_pages_per_second_upper_estimate": 0.368344894173716
+      "required_token_positions_per_second": 29062.176795601066,
+      "required_pages_per_second_upper_estimate": 0.3727783993997135
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.491669049414142,
-    "producer_p95_seconds": 0.2010451969999849,
-    "gpu_update_seconds": 0.9627902096648354,
-    "producer_batches_per_second": 2.033888448320206,
-    "gpu_batches_per_second": 1.038647869454466,
-    "producer_headroom": 1.958207885593098,
+    "producer_batch_seconds": 0.07545841072658277,
+    "producer_p95_seconds": 0.1775587700012693,
+    "gpu_update_seconds": 0.9513396121169038,
+    "producer_batches_per_second": 13.252333177588596,
+    "gpu_batches_per_second": 1.0511493343316358,
+    "producer_headroom": 12.607469504811109,
     "sustainable": true,
-    "buffer_seconds": 60.65578320888463,
+    "buffer_seconds": 60.88573517548184,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292133.8636338746,
-    "upload_bytes_per_second": 4497947.86412636,
+    "generation_bytes_per_second": 291825.79984629113,
+    "upload_bytes_per_second": 4568481.724434444,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.685126607534194,
+    "estimated_full_shards_per_hour": 62.61902329007674,
     "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": 1.5412007810000432,
-      "bytes": 6087939,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0659989769999356,
       "bytes": 6165937,
@@ -562,6 +556,12 @@
       "bytes": 2307263,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.328747977000603,
+      "bytes": 11901603,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -85,3 +85,4 @@
 {"step": 448700, "loss": 0.5940396189689636, "lr": 0.0003, "grad_norm": 0.15803821648863708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65676020290, "tokens_per_second": 152803.27240567782, "data_wait_s": 1.5137999071157537e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 448800, "loss": 0.5863661766052246, "lr": 0.0003, "grad_norm": 0.1476474418774333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65690699324, "tokens_per_second": 152705.65303955795, "data_wait_s": 1.2272999811102636e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 448900, "loss": 0.6093892455101013, "lr": 0.0003, "grad_norm": 0.1593004657932762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65705297294, "tokens_per_second": 152426.27186727803, "data_wait_s": 1.1653999536065385e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 449000, "loss": 0.5862565636634827, "lr": 0.0003, "grad_norm": 0.21102991124665044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65719775824, "tokens_per_second": 150423.05749013033, "data_wait_s": 1.5794999853824265e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4214a951d8bb152340d75c85ff91cb8f19c7584053ce093df782552999caf8d0
-size 185214
+oid sha256:f232c435127a5626434d258f188e6132047e27fd03156e86ed18a0d853d6d7b1
+size 187332