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": 77128.39637932065,
-      "required_pages_per_second_upper_estimate": 0.005335921424057228
+      "required_token_positions_per_second": 77366.64535383106,
+      "required_pages_per_second_upper_estimate": 0.005352404040927642
     },
     {
       "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": 9641.049547415081,
-      "required_pages_per_second_upper_estimate": 0.004036506117494373
+      "required_token_positions_per_second": 9670.830669228882,
+      "required_pages_per_second_upper_estimate": 0.00404897485129727
     },
     {
       "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": 19282.099094830162,
-      "required_pages_per_second_upper_estimate": 0.004015952738304637
+      "required_token_positions_per_second": 19341.661338457765,
+      "required_pages_per_second_upper_estimate": 0.004028357982890272
     },
     {
       "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": 28923.14864224524,
-      "required_pages_per_second_upper_estimate": 0.0017338985980117514
+      "required_token_positions_per_second": 29012.492007686644,
+      "required_pages_per_second_upper_estimate": 0.0017392545963505434
     },
     {
       "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": 28923.14864224524,
-      "required_pages_per_second_upper_estimate": 0.00224864758744866
+      "required_token_positions_per_second": 29012.492007686644,
+      "required_pages_per_second_upper_estimate": 0.002255593641131796
     },
     {
       "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": 28923.14864224524,
-      "required_pages_per_second_upper_estimate": 0.3709950955252657
+      "required_token_positions_per_second": 29012.492007686644,
+      "required_pages_per_second_upper_estimate": 0.3721410962877162
     }
   ],
   "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": 0.11027624320320228,
-    "producer_p95_seconds": 0.3144188839942217,
-    "gpu_update_seconds": 0.9559125232865292,
-    "producer_batches_per_second": 9.068136263558907,
-    "gpu_batches_per_second": 1.0461208276275045,
-    "producer_headroom": 8.668345017204674,
+    "producer_batch_seconds": 0.14034434461700585,
+    "producer_p95_seconds": 0.3258947289941716,
+    "gpu_update_seconds": 0.9529688105618389,
+    "producer_batches_per_second": 7.125331645738632,
+    "gpu_batches_per_second": 1.0493522861576479,
+    "producer_headroom": 6.790218823298175,
     "sustainable": true,
-    "buffer_seconds": 60.22248896705134,
+    "buffer_seconds": 60.03703506539585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 273446.6088885254,
-    "upload_bytes_per_second": 3820036.469901518,
+    "generation_bytes_per_second": 295447.49217449536,
+    "upload_bytes_per_second": 3846108.1721407576,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.675276756208625,
+    "estimated_full_shards_per_hour": 63.39615415502687,
     "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": 1.3136005369960913,
-      "bytes": 2685859,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3078916630038293,
       "bytes": 5962936,
@@ -562,6 +556,12 @@
       "bytes": 12094685,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.475563729996793,
+      "bytes": 6268574,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -389,3 +389,4 @@
 {"step": 523900, "loss": 0.6106423735618591, "lr": 0.0003, "grad_norm": 0.19980792504345476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76681058358, "tokens_per_second": 153079.29437360534, "data_wait_s": 1.4077995729167014e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 524000, "loss": 0.5886540412902832, "lr": 0.0003, "grad_norm": 0.16838592874414218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76695771128, "tokens_per_second": 153939.2906792162, "data_wait_s": 1.3355995179153979e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 524100, "loss": 0.5677302479743958, "lr": 0.0003, "grad_norm": 0.13734187044656046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76710356884, "tokens_per_second": 152362.16531835718, "data_wait_s": 1.2313001207076013e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 524200, "loss": 0.5946225523948669, "lr": 0.0003, "grad_norm": 0.1904968002587605, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76725083929, "tokens_per_second": 153597.80008041637, "data_wait_s": 1.351499668089673e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:517f6d1a6c64c19294a3c50cc315238d6089518591a69b2cb98583742a32df97
-size 833784
+oid sha256:0abddeef33d7f9899d8c37b652c79d012f45fe641af26de2f9b763ba52c0d7a9
+size 835902