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": 77634.63562570863,
-      "required_pages_per_second_upper_estimate": 0.005370944229759234
+      "required_token_positions_per_second": 77249.15126579156,
+      "required_pages_per_second_upper_estimate": 0.0053442755272932255
     },
     {
       "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": 9704.329453213579,
-      "required_pages_per_second_upper_estimate": 0.004063000092617506
+      "required_token_positions_per_second": 9656.143908223945,
+      "required_pages_per_second_upper_estimate": 0.004042825811158948
     },
     {
       "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": 19408.658906427157,
-      "required_pages_per_second_upper_estimate": 0.004042311809453614
+      "required_token_positions_per_second": 19312.28781644789,
+      "required_pages_per_second_upper_estimate": 0.0040222402528874825
     },
     {
       "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": 29112.988359640738,
-      "required_pages_per_second_upper_estimate": 0.0017452791992011469
+      "required_token_positions_per_second": 28968.43172467184,
+      "required_pages_per_second_upper_estimate": 0.0017366132496599625
     },
     {
       "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": 29112.988359640738,
-      "required_pages_per_second_upper_estimate": 0.0022634067904594903
+      "required_token_positions_per_second": 28968.43172467184,
+      "required_pages_per_second_upper_estimate": 0.0022521681479280983
     },
     {
       "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": 29112.988359640738,
-      "required_pages_per_second_upper_estimate": 0.3734301555860076
+      "required_token_positions_per_second": 28968.43172467184,
+      "required_pages_per_second_upper_estimate": 0.3715759382854483
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14636889871093217,
-    "producer_p95_seconds": 0.22710263699991629,
-    "gpu_update_seconds": 0.9496792173464526,
-    "producer_batches_per_second": 6.83205249753861,
-    "gpu_batches_per_second": 1.0529871368504318,
-    "producer_headroom": 6.488258268732344,
+    "producer_batch_seconds": 0.1278357931329026,
+    "producer_p95_seconds": 0.23634675100038294,
+    "gpu_update_seconds": 0.9544182530410422,
+    "producer_batches_per_second": 7.8225352656932685,
+    "gpu_batches_per_second": 1.0477586705972164,
+    "producer_headroom": 7.465970442634914,
     "sustainable": true,
-    "buffer_seconds": 60.77946991017296,
+    "buffer_seconds": 60.12834994158566,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315616.77666087373,
-    "upload_bytes_per_second": 4778955.361641659,
+    "generation_bytes_per_second": 286499.0134102965,
+    "upload_bytes_per_second": 4771107.809323231,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.72401308889064,
+    "estimated_full_shards_per_hour": 61.47601892215416,
     "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": 1.2044296969997959,
-      "bytes": 2640995,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.348795689000326,
       "bytes": 5804474,
@@ -562,6 +556,12 @@
       "bytes": 11516865,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.9571627120021731,
+      "bytes": 762513,
+      "pages": 23,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -214,3 +214,4 @@
 {"step": 946400, "loss": 0.5355998873710632, "lr": 3e-05, "grad_norm": 0.3428753670861612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138526938165, "tokens_per_second": 153493.54854810852, "data_wait_s": 1.7089001630665734e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 946500, "loss": 0.5369380116462708, "lr": 3e-05, "grad_norm": 0.3448912691610507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138541491490, "tokens_per_second": 152682.47314526647, "data_wait_s": 1.62110009114258e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 946600, "loss": 0.4609610140323639, "lr": 3e-05, "grad_norm": 0.3421678232833132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138556086566, "tokens_per_second": 144029.58318641555, "data_wait_s": 9.654999303165823e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 946700, "loss": 0.5465481877326965, "lr": 3e-05, "grad_norm": 0.34224239445298343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138570649993, "tokens_per_second": 151297.40515588396, "data_wait_s": 1.3739001587964594e-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 sha256:7267487cf24d064d813803dd1dcf5ebd152c7b9ed58d2d524c4bd579a622456f
-size 460785
+oid sha256:551766bc65be54b526899349be3ea195ddffb92f635a47d25a09d327b26d51c2
+size 462903