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": 76431.22197205317,
-      "required_pages_per_second_upper_estimate": 0.005287689280894968
+      "required_token_positions_per_second": 76281.05800507293,
+      "required_pages_per_second_upper_estimate": 0.005277300589230868
     },
     {
       "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": 9553.902746506647,
-      "required_pages_per_second_upper_estimate": 0.0040000195717861545
+      "required_token_positions_per_second": 9535.132250634117,
+      "required_pages_per_second_upper_estimate": 0.003992160757136853
     },
     {
       "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": 19107.805493013293,
-      "required_pages_per_second_upper_estimate": 0.003979651977477559
+      "required_token_positions_per_second": 19070.264501268233,
+      "required_pages_per_second_upper_estimate": 0.003971833178919592
     },
     {
       "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": 28661.70823951994,
-      "required_pages_per_second_upper_estimate": 0.0017182256450647504
+      "required_token_positions_per_second": 28605.396751902346,
+      "required_pages_per_second_upper_estimate": 0.0017148498573647387
     },
     {
       "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": 28661.70823951994,
-      "required_pages_per_second_upper_estimate": 0.0022283217460915684
+      "required_token_positions_per_second": 28605.396751902346,
+      "required_pages_per_second_upper_estimate": 0.002223943775617358
     },
     {
       "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": 28661.70823951994,
-      "required_pages_per_second_upper_estimate": 0.3676416187519393
+      "required_token_positions_per_second": 28605.396751902346,
+      "required_pages_per_second_upper_estimate": 0.3669193154513455
     }
   ],
   "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": 0.11312911124201719,
-    "producer_p95_seconds": 0.2803567499977362,
-    "gpu_update_seconds": 0.9646319671162448,
-    "producer_batches_per_second": 8.839457757788791,
-    "gpu_batches_per_second": 1.0366647945428218,
-    "producer_headroom": 8.526823525136752,
+    "producer_batch_seconds": 0.10162633339851368,
+    "producer_p95_seconds": 0.2952909859995998,
+    "gpu_update_seconds": 0.9665309046329281,
+    "producer_batches_per_second": 9.8399692929847,
+    "gpu_batches_per_second": 1.0346280653899866,
+    "producer_headroom": 9.510634422308735,
     "sustainable": true,
-    "buffer_seconds": 60.77181392832342,
+    "buffer_seconds": 60.891446991874474,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284971.8245143501,
-    "upload_bytes_per_second": 4833244.433336547,
+    "generation_bytes_per_second": 297148.27298900066,
+    "upload_bytes_per_second": 4709981.621861987,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.14831973622205,
+    "estimated_full_shards_per_hour": 63.761102125668664,
     "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": 1.2431204950007668,
-      "bytes": 13191953,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.2758062670000072,
       "bytes": 1083353,
@@ -562,6 +556,12 @@
       "bytes": 7252352,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0232120440014114,
+      "bytes": 1138613,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 868700, "loss": 0.5076080560684204, "lr": 3e-05, "grad_norm": 0.3227668576759112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127153617932, "tokens_per_second": 150460.5379944936, "data_wait_s": 1.3911998394178227e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 868800, "loss": 0.5246954560279846, "lr": 3e-05, "grad_norm": 0.3131322136905751, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127168232000, "tokens_per_second": 151019.21597509962, "data_wait_s": 9.941999451257288e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 868900, "loss": 0.524967610836029, "lr": 3e-05, "grad_norm": 0.3019740572453568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127182912800, "tokens_per_second": 152499.50259307757, "data_wait_s": 1.1255000572418794e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 869000, "loss": 0.49112239480018616, "lr": 3e-05, "grad_norm": 0.29584658459382734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127197639142, "tokens_per_second": 153049.0122783494, "data_wait_s": 1.383099879603833e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd472026fec89453ec6dcf479158280bbb2201c538a42dbda514bc36f8d40b7d
-size 563232
+oid sha256:685e191446d8e319563a70ca81981c56d0063b9d683df820adfc12460dba1449
+size 565350