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": 76961.15338502685,
-      "required_pages_per_second_upper_estimate": 0.005324351165654773
+      "required_token_positions_per_second": 77112.18608782446,
+      "required_pages_per_second_upper_estimate": 0.005334799958478464
     },
     {
       "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": 9620.144173128356,
-      "required_pages_per_second_upper_estimate": 0.004027753473834379
+      "required_token_positions_per_second": 9639.023260978058,
+      "required_pages_per_second_upper_estimate": 0.004035657753676865
     },
     {
       "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": 19240.28834625671,
-      "required_pages_per_second_upper_estimate": 0.004007244662000375
+      "required_token_positions_per_second": 19278.046521956116,
+      "required_pages_per_second_upper_estimate": 0.004015108694248506
     },
     {
       "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": 28860.432519385067,
-      "required_pages_per_second_upper_estimate": 0.0017301388622082588
+      "required_token_positions_per_second": 28917.069782934173,
+      "required_pages_per_second_upper_estimate": 0.0017335341796779337
     },
     {
       "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": 28860.432519385067,
-      "required_pages_per_second_upper_estimate": 0.002243771684755352
+      "required_token_positions_per_second": 28917.069782934173,
+      "required_pages_per_second_upper_estimate": 0.002248174982875301
     },
     {
       "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": 28860.432519385067,
-      "required_pages_per_second_upper_estimate": 0.3701906404405417
+      "required_token_positions_per_second": 28917.069782934173,
+      "required_pages_per_second_upper_estimate": 0.370917122444994
     }
   ],
   "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": 0.08259558439064563,
-    "producer_p95_seconds": 0.2864679660015099,
-    "gpu_update_seconds": 0.9579898007914227,
-    "producer_batches_per_second": 12.107184753005454,
-    "gpu_batches_per_second": 1.0438524493411845,
-    "producer_headroom": 11.598559509676646,
+    "producer_batch_seconds": 0.09827977441398161,
+    "producer_p95_seconds": 0.29524791999938316,
+    "gpu_update_seconds": 0.95611347233795,
+    "producer_batches_per_second": 10.175033530171968,
+    "gpu_batches_per_second": 1.0459009614776538,
+    "producer_headroom": 9.72848663968779,
     "sustainable": true,
-    "buffer_seconds": 60.35335744985963,
+    "buffer_seconds": 60.23514875729085,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380606.0596187105,
-    "upload_bytes_per_second": 5335861.056249346,
+    "generation_bytes_per_second": 381558.63978710293,
+    "upload_bytes_per_second": 5303629.925148026,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.66920033856378,
+    "estimated_full_shards_per_hour": 81.8736018677694,
     "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": 1.1309305639988452,
-      "bytes": 2518678,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.3960003290012537,
       "bytes": 5982392,
@@ -562,6 +556,12 @@
       "bytes": 12022604,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.3143507820022933,
+      "bytes": 918492,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -294,3 +294,4 @@
 {"step": 633800, "loss": 0.5685456395149231, "lr": 0.0001, "grad_norm": 0.18621276709977438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92771598291, "tokens_per_second": 151822.60911483734, "data_wait_s": 1.2259999493835494e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 633900, "loss": 0.5527037382125854, "lr": 0.0001, "grad_norm": 0.17776146793990177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92786218479, "tokens_per_second": 151091.99041596107, "data_wait_s": 1.7308997485088184e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 634000, "loss": 0.5494862198829651, "lr": 0.0001, "grad_norm": 0.17554819351669215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92800778982, "tokens_per_second": 150739.39576523265, "data_wait_s": 1.6712001524865627e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 634100, "loss": 0.5661565065383911, "lr": 0.0001, "grad_norm": 0.16171603905457355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92815519092, "tokens_per_second": 153121.16051451332, "data_wait_s": 1.121400055126287e-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 sha256:1d84a7cd603818cc73d1b7c13421fa07e15507e3beae9efa0a66ea156824fb4f
-size 631791
+oid sha256:bd545aa2766de04a2bfdb88cad2a5160a6c2896c48f17b9fafd9cd43623c2d36
+size 633909