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": 77213.8692519842,
-      "required_pages_per_second_upper_estimate": 0.005341834635712481
+      "required_token_positions_per_second": 77086.67688126446,
+      "required_pages_per_second_upper_estimate": 0.005333035172379129
     },
     {
       "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": 9651.733656498025,
-      "required_pages_per_second_upper_estimate": 0.004040979330857833
+      "required_token_positions_per_second": 9635.834610158057,
+      "required_pages_per_second_upper_estimate": 0.004034322732165131
     },
     {
       "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": 19303.46731299605,
-      "required_pages_per_second_upper_estimate": 0.004020403174630783
+      "required_token_positions_per_second": 19271.669220316115,
+      "required_pages_per_second_upper_estimate": 0.0040137804704976476
     },
     {
       "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": 28955.20096949408,
-      "required_pages_per_second_upper_estimate": 0.0017358200860892493
+      "required_token_positions_per_second": 28907.503830474172,
+      "required_pages_per_second_upper_estimate": 0.0017329607154343203
     },
     {
       "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": 28955.20096949408,
-      "required_pages_per_second_upper_estimate": 0.0022511395149089693
+      "required_token_positions_per_second": 28907.503830474172,
+      "required_pages_per_second_upper_estimate": 0.002247431272147714
     },
     {
       "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": 28955.20096949408,
-      "required_pages_per_second_upper_estimate": 0.37140622836410614
+      "required_token_positions_per_second": 28907.503830474172,
+      "required_pages_per_second_upper_estimate": 0.3707944206779566
     }
   ],
   "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": 0.3253820029297003,
-    "producer_p95_seconds": 0.30500457099697087,
-    "gpu_update_seconds": 0.9548543637852389,
-    "producer_batches_per_second": 3.0733107270719358,
-    "gpu_batches_per_second": 1.047280127658206,
-    "producer_headroom": 2.934564159012623,
+    "producer_batch_seconds": 0.5078379566952549,
+    "producer_p95_seconds": 0.3021672039976693,
+    "gpu_update_seconds": 0.956429865481972,
+    "producer_batches_per_second": 1.9691320564289434,
+    "gpu_batches_per_second": 1.0455549707202754,
+    "producer_headroom": 1.8833367078465733,
     "sustainable": true,
-    "buffer_seconds": 60.15582491847005,
+    "buffer_seconds": 60.25508152536423,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332893.8081374496,
-    "upload_bytes_per_second": 4799437.4227447575,
+    "generation_bytes_per_second": 301560.6285065212,
+    "upload_bytes_per_second": 4754873.459337372,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.43126185505501,
+    "estimated_full_shards_per_hour": 64.70789090534905,
     "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": 1.2765829149975616,
-      "bytes": 9916034,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.591446239999641,
       "bytes": 5814050,
@@ -562,6 +556,12 @@
       "bytes": 6176798,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0939249449947965,
+      "bytes": 5048949,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -362,3 +362,4 @@
 {"step": 796200, "loss": 0.5428974628448486, "lr": 3e-05, "grad_norm": 0.2584630178822616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116541001292, "tokens_per_second": 151880.1621858557, "data_wait_s": 1.971399615285918e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 796300, "loss": 0.4854239225387573, "lr": 3e-05, "grad_norm": 0.2590141382661817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116555675906, "tokens_per_second": 152579.68360035724, "data_wait_s": 1.1572999937925488e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 796400, "loss": 0.5987265110015869, "lr": 3e-05, "grad_norm": 0.26447540313313356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116570295974, "tokens_per_second": 135728.0809670561, "data_wait_s": 1.1076997907366604e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 796500, "loss": 0.4834693968296051, "lr": 3e-05, "grad_norm": 0.26357056321213773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116585015541, "tokens_per_second": 153412.46184235357, "data_wait_s": 1.1577998520806432e-05, "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 sha256:6f31d5916526812b7d66d6f19f79c57190d6462dfc8ef7cd82786352603282c6
-size 776598
+oid sha256:feaf59ff365802e1da7b78a8177578cbefb733a7e51a0f621086ebd70c932c65
+size 778716