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": 75845.4656511612,
-      "required_pages_per_second_upper_estimate": 0.005247165299473743
+      "required_token_positions_per_second": 76136.53466826682,
+      "required_pages_per_second_upper_estimate": 0.005267302129450278
     },
     {
       "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": 9480.68320639515,
-      "required_pages_per_second_upper_estimate": 0.003969364079339337
+      "required_token_positions_per_second": 9517.066833533352,
+      "required_pages_per_second_upper_estimate": 0.003984597144245571
     },
     {
       "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": 18961.3664127903,
-      "required_pages_per_second_upper_estimate": 0.003949152578925354
+      "required_token_positions_per_second": 19034.133667066704,
+      "required_pages_per_second_upper_estimate": 0.003964308078989586
     },
     {
       "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": 28442.04961918545,
-      "required_pages_per_second_upper_estimate": 0.0017050574461749907
+      "required_token_positions_per_second": 28551.200500600055,
+      "required_pages_per_second_upper_estimate": 0.0017116008748520504
     },
     {
       "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": 28442.04961918545,
-      "required_pages_per_second_upper_estimate": 0.002211244254536724
+      "required_token_positions_per_second": 28551.200500600055,
+      "required_pages_per_second_upper_estimate": 0.0022197302554627213
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28442.04961918545,
-      "required_pages_per_second_upper_estimate": 0.3648240738213395
+      "required_token_positions_per_second": 28551.200500600055,
+      "required_pages_per_second_upper_estimate": 0.3662241441310406
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1404301015938927,
-    "producer_p95_seconds": 0.2638170690006518,
-    "gpu_update_seconds": 0.972081842559974,
-    "producer_batches_per_second": 7.120980392735755,
-    "gpu_batches_per_second": 1.028719965971696,
-    "producer_headroom": 6.92217574100402,
+    "producer_batch_seconds": 0.08798861610148379,
+    "producer_p95_seconds": 0.2519642240004032,
+    "gpu_update_seconds": 0.9683655858680593,
+    "producer_batches_per_second": 11.365106582044953,
+    "gpu_batches_per_second": 1.0326678421802682,
+    "producer_headroom": 11.005578093774899,
     "sustainable": true,
-    "buffer_seconds": 60.26907423871839,
+    "buffer_seconds": 60.038666323819676,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266300.58679623273,
-    "upload_bytes_per_second": 4984750.1514002085,
+    "generation_bytes_per_second": 265586.82723792625,
+    "upload_bytes_per_second": 4964788.564340346,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.14190676608311,
+    "estimated_full_shards_per_hour": 56.98875058034268,
     "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": 1.4511640289983916,
-      "bytes": 7595757,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2930020470012096,
       "bytes": 547633,
@@ -562,6 +556,12 @@
       "bytes": 12006565,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.637713142998109,
+      "bytes": 6812710,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -177,3 +177,4 @@
 {"step": 737700, "loss": 0.5151412487030029, "lr": 5e-05, "grad_norm": 0.22466648249232482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107978785707, "tokens_per_second": 149306.4756886534, "data_wait_s": 2.1768999431515113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737800, "loss": 0.6282327175140381, "lr": 5e-05, "grad_norm": 0.27595751261107293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107993394586, "tokens_per_second": 149611.35810698985, "data_wait_s": 2.9964001441840082e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737900, "loss": 0.5347310304641724, "lr": 5e-05, "grad_norm": 0.2168170256007732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108008061585, "tokens_per_second": 150672.47344936585, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 738000, "loss": 0.4385727047920227, "lr": 5e-05, "grad_norm": 0.22462039073004075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108022860653, "tokens_per_second": 152859.93455801453, "data_wait_s": 1.8253998860018328e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5942470efc3dcdf644427e468ab55cacdab28505155b66ef7166ed68be25ee0c
-size 381636
+oid sha256:084bb3bffa732f9497da406cc5726e564c0fae10ff9743343bbe715c7520d3f1
+size 383754