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": 77135.08397927361,
-      "required_pages_per_second_upper_estimate": 0.005336384087739344
+      "required_token_positions_per_second": 76847.30726777944,
+      "required_pages_per_second_upper_estimate": 0.0053164750271042106
     },
     {
       "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": 9641.885497409201,
-      "required_pages_per_second_upper_estimate": 0.004036856112300291
+      "required_token_positions_per_second": 9605.91340847243,
+      "required_pages_per_second_upper_estimate": 0.0040217953498451016
     },
     {
       "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": 19283.770994818402,
-      "required_pages_per_second_upper_estimate": 0.0040163009509812195
+      "required_token_positions_per_second": 19211.82681694486,
+      "required_pages_per_second_upper_estimate": 0.0040013168760257185
     },
     {
       "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": 28925.6564922276,
-      "required_pages_per_second_upper_estimate": 0.0017340489397889287
+      "required_token_positions_per_second": 28817.74022541729,
+      "required_pages_per_second_upper_estimate": 0.0017275795243722489
     },
     {
       "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": 28925.6564922276,
-      "required_pages_per_second_upper_estimate": 0.002248842561753922
+      "required_token_positions_per_second": 28817.74022541729,
+      "required_pages_per_second_upper_estimate": 0.002240452546682913
     },
     {
       "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": 28925.6564922276,
-      "required_pages_per_second_upper_estimate": 0.3710272635321199
+      "required_token_positions_per_second": 28817.74022541729,
+      "required_pages_per_second_upper_estimate": 0.3696430295329368
     }
   ],
   "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": 0.0955585915000654,
-    "producer_p95_seconds": 0.24718180399941048,
-    "gpu_update_seconds": 0.9558296458173418,
-    "producer_batches_per_second": 10.464783797062513,
-    "gpu_batches_per_second": 1.0462115340070748,
-    "producer_headroom": 10.002550590301318,
+    "producer_batch_seconds": 0.07435815143773539,
+    "producer_p95_seconds": 0.25384056300026714,
+    "gpu_update_seconds": 0.9594090231826861,
+    "producer_batches_per_second": 13.44842469406143,
+    "gpu_batches_per_second": 1.0423083125512618,
+    "producer_headroom": 12.90253999907539,
     "sustainable": true,
-    "buffer_seconds": 60.217267686492534,
+    "buffer_seconds": 60.44276846050922,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330392.75470309873,
-    "upload_bytes_per_second": 4839515.31852037,
+    "generation_bytes_per_second": 378501.4766605139,
+    "upload_bytes_per_second": 4871900.967954124,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.8945940095875,
+    "estimated_full_shards_per_hour": 81.21760582791865,
     "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": 0.843333258999337,
-      "bytes": 1195587,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.4843279429987888,
       "bytes": 7415076,
@@ -562,6 +556,12 @@
       "bytes": 7186634,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.2788624959976005,
+      "bytes": 6210071,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 948900, "loss": 0.5397599339485168, "lr": 3e-05, "grad_norm": 0.35230049628193194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138892548315, "tokens_per_second": 150450.3322366843, "data_wait_s": 1.689300188445486e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949000, "loss": 0.5276049375534058, "lr": 3e-05, "grad_norm": 0.35546142067112296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138907098420, "tokens_per_second": 151543.5062288765, "data_wait_s": 1.0994001058861613e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949100, "loss": 0.5535507202148438, "lr": 3e-05, "grad_norm": 0.37402986916779357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138921774854, "tokens_per_second": 153016.35622632026, "data_wait_s": 1.4047000149730593e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 949200, "loss": 0.5632185339927673, "lr": 3e-05, "grad_norm": 0.35542638318195346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138936514089, "tokens_per_second": 153708.60155259547, "data_wait_s": 1.0625997674651444e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8d1bff6f8c9cc30f6c75f519e40a899868699ce95963d555217408483d74c645
-size 513735
+oid sha256:1d781be6d5a8919441e78413ed3d5794c8fbb276e4d206c8693a82a16598f512
+size 515853