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": 76431.90732465035,
-      "required_pages_per_second_upper_estimate": 0.005287736695177878
+      "required_token_positions_per_second": 76409.77180995035,
+      "required_pages_per_second_upper_estimate": 0.005286205308909462
     },
     {
       "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": 9553.988415581294,
-      "required_pages_per_second_upper_estimate": 0.004000055439638736
+      "required_token_positions_per_second": 9551.221476243794,
+      "required_pages_per_second_upper_estimate": 0.003998896979918389
     },
     {
       "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": 19107.97683116259,
-      "required_pages_per_second_upper_estimate": 0.003979687662695567
+      "required_token_positions_per_second": 19102.44295248759,
+      "required_pages_per_second_upper_estimate": 0.0039785351017057585
     },
     {
       "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": 28661.965246743883,
-      "required_pages_per_second_upper_estimate": 0.0017182410522553985
+      "required_token_positions_per_second": 28653.664428731383,
+      "required_pages_per_second_upper_estimate": 0.0017177434308901644
     },
     {
       "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": 28661.965246743883,
-      "required_pages_per_second_upper_estimate": 0.0022283417272727746
+      "required_token_positions_per_second": 28653.664428731383,
+      "required_pages_per_second_upper_estimate": 0.0022276963751837425
     },
     {
       "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": 28661.965246743883,
-      "required_pages_per_second_upper_estimate": 0.36764491536465516
+      "required_token_positions_per_second": 28653.664428731383,
+      "required_pages_per_second_upper_estimate": 0.36753844138391484
     }
   ],
   "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": 0.2512014338753943,
+    "producer_batch_seconds": 0.1449134782031365,
     "producer_p95_seconds": 0.3332882180038723,
-    "gpu_update_seconds": 0.9646233174168308,
-    "producer_batches_per_second": 3.980868996536218,
-    "gpu_batches_per_second": 1.0366740902323455,
-    "producer_headroom": 3.840039057640577,
+    "gpu_update_seconds": 0.9649027637902051,
+    "producer_batches_per_second": 6.900669367677603,
+    "gpu_batches_per_second": 1.03637385809937,
+    "producer_headroom": 6.658474944874526,
     "sustainable": true,
-    "buffer_seconds": 60.77126899726034,
+    "buffer_seconds": 60.78887411878292,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 160,
+    "upload_pages_per_commit": 163,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 645379.1960824085,
-    "upload_bytes_per_second": 4452799.972175803,
+    "generation_bytes_per_second": 616913.9635945064,
+    "upload_bytes_per_second": 4499680.571279526,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.24167591025444,
+    "estimated_full_shards_per_hour": 66.18768778265186,
     "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": 1.3224051930010319,
-      "bytes": 6398826,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.916404732997762,
       "bytes": 5130348,
@@ -562,6 +556,12 @@
       "bytes": 5830481,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.2900278290035203,
+      "bytes": 10699576,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -431,3 +431,4 @@
 {"step": 763100, "loss": 0.5360861420631409, "lr": 5e-05, "grad_norm": 0.2544982764437836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111695162754, "tokens_per_second": 150410.954031996, "data_wait_s": 1.286400220124051e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 763200, "loss": 0.5217152833938599, "lr": 5e-05, "grad_norm": 0.23616950554153632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111709830959, "tokens_per_second": 151695.11723153258, "data_wait_s": 1.3425000361166894e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 763300, "loss": 0.5940197706222534, "lr": 5e-05, "grad_norm": 0.2692685291579838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111724509274, "tokens_per_second": 151803.64970920098, "data_wait_s": 1.1246003850828856e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 763400, "loss": 0.5920512676239014, "lr": 5e-05, "grad_norm": 0.2569159887800782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111739120005, "tokens_per_second": 150828.2406486852, "data_wait_s": 1.436899765394628e-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 sha256:6187685816537f2d5a9dfa0b5b7c5918c5d1e0bd6550e6848ef19bfbd3e2370f
-size 923523
+oid sha256:c12e1e4147fc4d3f871f58a81ed1159ee040f90b27d319f4db3f8e1a0b75fb5b
+size 925641