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": 76891.52129667504,
-      "required_pages_per_second_upper_estimate": 0.005319533856213889
+      "required_token_positions_per_second": 77362.01015544092,
+      "required_pages_per_second_upper_estimate": 0.00535208336714787
     },
     {
       "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": 9611.44016208438,
-      "required_pages_per_second_upper_estimate": 0.004024109286170686
+      "required_token_positions_per_second": 9670.251269430115,
+      "required_pages_per_second_upper_estimate": 0.004048732268700769
     },
     {
       "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": 19222.88032416876,
-      "required_pages_per_second_upper_estimate": 0.004003619030079869
+      "required_token_positions_per_second": 19340.50253886023,
+      "required_pages_per_second_upper_estimate": 0.004028116635493706
     },
     {
       "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": 28834.32048625314,
-      "required_pages_per_second_upper_estimate": 0.0017285734856927657
+      "required_token_positions_per_second": 29010.753808290345,
+      "required_pages_per_second_upper_estimate": 0.0017391503939508121
     },
     {
       "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": 28834.32048625314,
-      "required_pages_per_second_upper_estimate": 0.0022417415890340405
+      "required_token_positions_per_second": 29010.753808290345,
+      "required_pages_per_second_upper_estimate": 0.0022554585037742646
     },
     {
       "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": 28834.32048625314,
-      "required_pages_per_second_upper_estimate": 0.369855703316442
+      "required_token_positions_per_second": 29010.753808290345,
+      "required_pages_per_second_upper_estimate": 0.3721188005321936
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04985304220309672,
-    "producer_p95_seconds": 0.1690451349995783,
-    "gpu_update_seconds": 0.9588573454741643,
-    "producer_batches_per_second": 20.05895640081686,
-    "gpu_batches_per_second": 1.0429080036983918,
-    "producer_headroom": 19.233677687469253,
+    "producer_batch_seconds": 0.06796342360943441,
+    "producer_p95_seconds": 0.18086238599971693,
+    "gpu_update_seconds": 0.9530259083477896,
+    "producer_batches_per_second": 14.713796729056833,
+    "gpu_batches_per_second": 1.0492894172558718,
+    "producer_headroom": 14.022629492954122,
     "sustainable": true,
-    "buffer_seconds": 60.40801276487235,
+    "buffer_seconds": 60.04063222591074,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412789.11763402086,
-    "upload_bytes_per_second": 5088797.081617949,
+    "generation_bytes_per_second": 433808.4152634382,
+    "upload_bytes_per_second": 4929107.460167297,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.57493540540189,
+    "estimated_full_shards_per_hour": 93.08518737246855,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 7239835,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0533643100006884,
+      "bytes": 1118487,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -59,3 +59,4 @@
 {"step": 765900, "loss": 0.5385651588439941, "lr": 3e-05, "grad_norm": 0.24259133436256383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112105413335, "tokens_per_second": 151001.37321612364, "data_wait_s": 1.3341999874683097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 766000, "loss": 0.6250334978103638, "lr": 3e-05, "grad_norm": 0.34156944122485927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112119933901, "tokens_per_second": 151121.8795573869, "data_wait_s": 1.476399938837858e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 766100, "loss": 0.49440351128578186, "lr": 3e-05, "grad_norm": 0.2436492609561353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112134464551, "tokens_per_second": 151654.22812500523, "data_wait_s": 9.52400023379596e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 766200, "loss": 0.564631998538971, "lr": 3e-05, "grad_norm": 0.280078313192073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112149023207, "tokens_per_second": 141144.55896482535, "data_wait_s": 9.847000001173e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:635a9b385767a2d552fa15a5225fafa41bb3eadf54cdbede0966da1e0eed2213
-size 130146
+oid sha256:884a5a7ab94e28b2966d193f3f8f9aa6ce676db4c2695f466a82067c398098b8
+size 132264