Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75521.86830531279,
-      "required_pages_per_second_upper_estimate": 0.0065460441362792336
+      "required_token_positions_per_second": 76056.2065065976,
+      "required_pages_per_second_upper_estimate": 0.006592359217298812
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9440.233538164099,
-      "required_pages_per_second_upper_estimate": 0.0037658503024429945
+      "required_token_positions_per_second": 9507.0258133247,
+      "required_pages_per_second_upper_estimate": 0.0037924947396380648
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18880.467076328197,
-      "required_pages_per_second_upper_estimate": 0.004367663773236922
+      "required_token_positions_per_second": 19014.0516266494,
+      "required_pages_per_second_upper_estimate": 0.004398566207945945
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28320.7006144923,
-      "required_pages_per_second_upper_estimate": 0.0017245820801898833
+      "required_token_positions_per_second": 28521.077439974102,
+      "required_pages_per_second_upper_estimate": 0.00173678397756577
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28320.7006144923,
-      "required_pages_per_second_upper_estimate": 0.0021802059300205312
+      "required_token_positions_per_second": 28521.077439974102,
+      "required_pages_per_second_upper_estimate": 0.0021956314927246744
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28320.7006144923,
-      "required_pages_per_second_upper_estimate": 0.35164831834426785
+      "required_token_positions_per_second": 28521.077439974102,
+      "required_pages_per_second_upper_estimate": 0.35413632789865657
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.0675434155233603,
-    "producer_p95_seconds": 0.2021248929995636,
-    "gpu_update_seconds": 0.9762470348580267,
-    "producer_batches_per_second": 14.805292155445471,
-    "gpu_batches_per_second": 1.0243308960681532,
-    "producer_headroom": 14.453622566960442,
+    "producer_batch_seconds": 0.08393217017172105,
+    "producer_p95_seconds": 0.25809267300064676,
+    "gpu_update_seconds": 0.969388342996102,
+    "producer_batches_per_second": 11.914382744471514,
+    "gpu_batches_per_second": 1.0315783217583225,
+    "producer_headroom": 11.54966374648459,
     "sustainable": true,
-    "buffer_seconds": 60.52731616119765,
+    "buffer_seconds": 60.102077265758325,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 557482.0799935845,
-    "upload_bytes_per_second": 6136706.52351664,
+    "generation_bytes_per_second": 561254.2974125833,
+    "upload_bytes_per_second": 6198286.288939025,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.62267684798863,
+    "estimated_full_shards_per_hour": 60.21605344668924,
     "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": 1.0273279060002096,
-      "bytes": 1056303,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1538401379993957,
       "bytes": 11570054,
@@ -562,6 +556,12 @@
       "bytes": 18374728,
       "pages": 40,
       "shards": 5
+    },
+    {
+      "seconds": 1.1437757880012214,
+      "bytes": 6056576,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 52600, "loss": 0.6517808940261602, "lr": 0.0003, "grad_norm": 0.14153193603161499, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7698377563, "tokens_per_second": 149626.31387750665, "data_wait_s": 5.992000296828337e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 52700, "loss": 0.6120903808623552, "lr": 0.0003, "grad_norm": 0.14055092120346682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7713113620, "tokens_per_second": 150284.65807365673, "data_wait_s": 1.345999953628052e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 52800, "loss": 0.6057971864938736, "lr": 0.0003, "grad_norm": 0.14566707485955477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7727824093, "tokens_per_second": 150764.41672000327, "data_wait_s": 1.2390000847517513e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 52900, "loss": 0.6748346481472254, "lr": 0.0003, "grad_norm": 0.12579535463662597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7742525265, "tokens_per_second": 150655.0856055337, "data_wait_s": 1.0423998901387677e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f0f5d3a05b263062d248648b32cc06424e281a67feece56ab51098ea9fb4bd05
-size 272835
+oid sha256:fe693a86d07245c51393becfcd5e076a4e2d78db8007d787ad008b3313737089
+size 274953