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": 76399.32048164976,
-      "required_pages_per_second_upper_estimate": 0.0052854822617672045
+      "required_token_positions_per_second": 77344.65022357054,
+      "required_pages_per_second_upper_estimate": 0.005350882366780486
     },
     {
       "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": 9549.91506020622,
-      "required_pages_per_second_upper_estimate": 0.003998350010804525
+      "required_token_positions_per_second": 9668.081277946318,
+      "required_pages_per_second_upper_estimate": 0.004047823738581079
     },
     {
       "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": 19099.83012041244,
-      "required_pages_per_second_upper_estimate": 0.0039779909176895206
+      "required_token_positions_per_second": 19336.162555892635,
+      "required_pages_per_second_upper_estimate": 0.004027212731494604
     },
     {
       "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": 28649.74518061866,
-      "required_pages_per_second_upper_estimate": 0.0017175084779501519
+      "required_token_positions_per_second": 29004.24383383895,
+      "required_pages_per_second_upper_estimate": 0.0017387601309226065
     },
     {
       "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": 28649.74518061866,
-      "required_pages_per_second_upper_estimate": 0.002227391670881929
+      "required_token_positions_per_second": 29004.24383383895,
+      "required_pages_per_second_upper_estimate": 0.0022549523818950206
     },
     {
       "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": 28649.74518061866,
-      "required_pages_per_second_upper_estimate": 0.3674881694772855
+      "required_token_positions_per_second": 29004.24383383895,
+      "required_pages_per_second_upper_estimate": 0.3720352975697971
     }
   ],
   "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": 0.05781445724994683,
-    "producer_p95_seconds": 0.20112377800069225,
-    "gpu_update_seconds": 0.9650347612412158,
-    "producer_batches_per_second": 17.296711714800708,
-    "gpu_batches_per_second": 1.0362321028869597,
-    "producer_headroom": 16.69192805995084,
+    "producer_batch_seconds": 0.1106320823202509,
+    "producer_p95_seconds": 0.22376580000127433,
+    "gpu_update_seconds": 0.9532398140903561,
+    "producer_batches_per_second": 9.038969338977656,
+    "gpu_batches_per_second": 1.0490539581104945,
+    "producer_headroom": 8.61630545225549,
     "sustainable": true,
-    "buffer_seconds": 60.7971899581966,
+    "buffer_seconds": 60.05410828769243,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 455695.0893840644,
-    "upload_bytes_per_second": 5105392.51173705,
+    "generation_bytes_per_second": 453688.19027739455,
+    "upload_bytes_per_second": 5199576.169823156,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.78155814305734,
+    "estimated_full_shards_per_hour": 97.35092431298615,
     "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": 1.5793915419999394,
-      "bytes": 781594,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.4397608880008193,
       "bytes": 13440043,
@@ -562,6 +556,12 @@
       "bytes": 10171575,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.4380360830000427,
+      "bytes": 7912633,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -156,3 +156,4 @@
 {"step": 620000, "loss": 0.519646942615509, "lr": 0.0001, "grad_norm": 0.15932970628983065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90751225678, "tokens_per_second": 154578.89700448365, "data_wait_s": 9.052999303094111e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 620100, "loss": 0.6086562871932983, "lr": 0.0001, "grad_norm": 0.1629300007082568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90765923478, "tokens_per_second": 141751.49508206395, "data_wait_s": 1.1692000043694861e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 620200, "loss": 0.542997419834137, "lr": 0.0001, "grad_norm": 0.15714926735442414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90780713195, "tokens_per_second": 154154.6075574205, "data_wait_s": 9.972998668672517e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 620300, "loss": 0.5748096108436584, "lr": 0.0001, "grad_norm": 0.1677302391892716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90795426171, "tokens_per_second": 153378.42680677114, "data_wait_s": 1.5035000615171157e-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 sha256:0ea90cc59ca8b707ff7144b1ac51e1dd71a941f12bb128bac19bfb0f16e44b1d
-size 337941
+oid sha256:a2738741e5dbea72da91522e5fed52c77eadb9694861a84887eb1925caf82eaf
+size 340059