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": 76314.58911435366,
-      "required_pages_per_second_upper_estimate": 0.005279620349173801
+      "required_token_positions_per_second": 77305.86623685164,
+      "required_pages_per_second_upper_estimate": 0.005348199200588025
     },
     {
       "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": 9539.323639294207,
-      "required_pages_per_second_upper_estimate": 0.003993915604042685
+      "required_token_positions_per_second": 9663.233279606455,
+      "required_pages_per_second_upper_estimate": 0.004045793982913896
     },
     {
       "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": 19078.647278588414,
-      "required_pages_per_second_upper_estimate": 0.003973579090366681
+      "required_token_positions_per_second": 19326.46655921291,
+      "required_pages_per_second_upper_estimate": 0.0040251933110868445
     },
     {
       "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": 28617.970917882623,
-      "required_pages_per_second_upper_estimate": 0.0017156036594156165
+      "required_token_positions_per_second": 28989.699838819364,
+      "required_pages_per_second_upper_estimate": 0.0017378882406285767
     },
     {
       "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": 28617.970917882623,
-      "required_pages_per_second_upper_estimate": 0.0022249213617144136
+      "required_token_positions_per_second": 28989.699838819364,
+      "required_pages_per_second_upper_estimate": 0.002253821650254521
     },
     {
       "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": 28617.970917882623,
-      "required_pages_per_second_upper_estimate": 0.3670806033514529
+      "required_token_positions_per_second": 28989.699838819364,
+      "required_pages_per_second_upper_estimate": 0.371848742817811
     }
   ],
   "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": 0.15585856931289754,
-    "producer_p95_seconds": 0.382340694006416,
-    "gpu_update_seconds": 0.9661062302192601,
-    "producer_batches_per_second": 6.416073266991348,
-    "gpu_batches_per_second": 1.035082860166472,
-    "producer_headroom": 6.198608356783583,
+    "producer_batch_seconds": 0.15348295925059574,
+    "producer_p95_seconds": 0.36698411800171016,
+    "gpu_update_seconds": 0.9537180499874397,
+    "producer_batches_per_second": 6.515381283255512,
+    "gpu_batches_per_second": 1.0485279166239643,
+    "producer_headroom": 6.21383673239111,
     "sustainable": true,
-    "buffer_seconds": 60.86469250381339,
+    "buffer_seconds": 60.084237149208704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457851.48374966515,
-    "upload_bytes_per_second": 4678131.604082665,
+    "generation_bytes_per_second": 439963.0909091533,
+    "upload_bytes_per_second": 4686631.055964475,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.24427017562357,
+    "estimated_full_shards_per_hour": 94.4058374925227,
     "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": 1.1959701189989573,
-      "bytes": 1136148,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.5574687859989353,
       "bytes": 11997417,
@@ -562,6 +556,12 @@
       "bytes": 5925879,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 2.3598094420012785,
+      "bytes": 7376391,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -390,3 +390,4 @@
 {"step": 643400, "loss": 0.5416357517242432, "lr": 0.0001, "grad_norm": 0.18411690164124123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94177104692, "tokens_per_second": 151851.7961083343, "data_wait_s": 1.3062999641988426e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 643500, "loss": 0.6674120426177979, "lr": 0.0001, "grad_norm": 0.2926909605467959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94191707692, "tokens_per_second": 150931.21682982807, "data_wait_s": 1.7291000403929502e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 643600, "loss": 0.5676829218864441, "lr": 0.0001, "grad_norm": 0.17735923865348308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94206449009, "tokens_per_second": 153685.1716380194, "data_wait_s": 1.135400088969618e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 643700, "loss": 0.5304334163665771, "lr": 0.0001, "grad_norm": 0.1813797392604267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94221115269, "tokens_per_second": 152512.29277982426, "data_wait_s": 1.2910000805277377e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3fe6d71487b49b2b401bedb28119cb042aa7f787e1bfed71b94c144b94b9262
-size 836685
+oid sha256:30d97d27b3796130bff7286e08d1197c784ad4c109eef1d23e663732cd1e7536
+size 838803