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": 76106.14576929317,
-      "required_pages_per_second_upper_estimate": 0.0052651997549073215
+      "required_token_positions_per_second": 76539.7048339045,
+      "required_pages_per_second_upper_estimate": 0.005295194376992762
     },
     {
       "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": 9513.268221161647,
-      "required_pages_per_second_upper_estimate": 0.003983006744569584
+      "required_token_positions_per_second": 9567.463104238062,
+      "required_pages_per_second_upper_estimate": 0.004005697010395829
     },
     {
       "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": 19026.536442323293,
-      "required_pages_per_second_upper_estimate": 0.003962725777427824
+      "required_token_positions_per_second": 19134.926208476125,
+      "required_pages_per_second_upper_estimate": 0.003985300507287104
     },
     {
       "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": 28539.80466348494,
-      "required_pages_per_second_upper_estimate": 0.0017109177118174352
+      "required_token_positions_per_second": 28702.38931271419,
+      "required_pages_per_second_upper_estimate": 0.0017206644132863435
     },
     {
       "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": 28539.80466348494,
-      "required_pages_per_second_upper_estimate": 0.002218844279252011
+      "required_token_positions_per_second": 28702.38931271419,
+      "required_pages_per_second_upper_estimate": 0.002231484520595291
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28539.80466348494,
-      "required_pages_per_second_upper_estimate": 0.3660779705684245
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28702.38931271419,
+      "required_pages_per_second_upper_estimate": 0.3681634318789419
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5060984954140508,
-    "producer_p95_seconds": 0.359134611004265,
-    "gpu_update_seconds": 0.9687522506197824,
-    "producer_batches_per_second": 1.9758999662345904,
-    "gpu_batches_per_second": 1.0322556663586857,
-    "producer_headroom": 1.9141575392893115,
+    "producer_batch_seconds": 0.11594883083631657,
+    "producer_p95_seconds": 0.36349311500089243,
+    "gpu_update_seconds": 0.9632647546785545,
+    "producer_batches_per_second": 8.624494035750017,
+    "gpu_batches_per_second": 1.0381361875258315,
+    "producer_headroom": 8.307671131573397,
     "sustainable": true,
-    "buffer_seconds": 60.06263953842651,
+    "buffer_seconds": 60.685679544748936,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285118.1376276385,
-    "upload_bytes_per_second": 4334084.972636399,
+    "generation_bytes_per_second": 273455.2642909847,
+    "upload_bytes_per_second": 4445335.933538257,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.179715124338784,
+    "estimated_full_shards_per_hour": 58.67713400408894,
     "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": 1.4634708860030514,
-      "bytes": 987981,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.6518487119974452,
       "bytes": 7534957,
@@ -562,6 +556,12 @@
       "bytes": 2867400,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.2084345140028745,
+      "bytes": 10491799,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -418,3 +418,4 @@
 {"step": 841800, "loss": 0.5760037899017334, "lr": 3e-05, "grad_norm": 0.32608737617193206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123215626343, "tokens_per_second": 149530.09190303768, "data_wait_s": 1.3249999028630555e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 841900, "loss": 0.5589576363563538, "lr": 3e-05, "grad_norm": 0.3204468003975533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123230326054, "tokens_per_second": 129672.86882570645, "data_wait_s": 1.219999830937013e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 842000, "loss": 0.5624033212661743, "lr": 3e-05, "grad_norm": 0.28762307969362577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123244957115, "tokens_per_second": 151700.78051702792, "data_wait_s": 1.1224001355003566e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 842100, "loss": 0.5551508665084839, "lr": 3e-05, "grad_norm": 0.31770559703554857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123259666338, "tokens_per_second": 152730.96904519544, "data_wait_s": 1.0275995009578764e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c79c1907354ae16c2a5fb4bd673274ff36c4f3e6e4ea8f4ff03d104c82cc3b86
-size 895989
+oid sha256:be8847e2a05351487a5663e1fc403d5d7c5868b75df96584a320effdc7aeb080
+size 898107