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": 76649.61994722096,
-      "required_pages_per_second_upper_estimate": 0.005302798559570195
+      "required_token_positions_per_second": 76634.62956973384,
+      "required_pages_per_second_upper_estimate": 0.005301761490473168
     },
     {
       "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": 9581.20249340262,
-      "required_pages_per_second_upper_estimate": 0.004011449405727958
+      "required_token_positions_per_second": 9579.32869621673,
+      "required_pages_per_second_upper_estimate": 0.00401066488597556
     },
     {
       "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": 19162.40498680524,
-      "required_pages_per_second_upper_estimate": 0.0039910236121489424
+      "required_token_positions_per_second": 19158.65739243346,
+      "required_pages_per_second_upper_estimate": 0.00399024308707201
     },
     {
       "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": 28743.60748020786,
-      "required_pages_per_second_upper_estimate": 0.0017231353794910913
+      "required_token_positions_per_second": 28737.98608865019,
+      "required_pages_per_second_upper_estimate": 0.001722798385650577
     },
     {
       "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": 28743.60748020786,
-      "required_pages_per_second_upper_estimate": 0.0022346890518183653
+      "required_token_positions_per_second": 28737.98608865019,
+      "required_pages_per_second_upper_estimate": 0.0022342520133506447
     },
     {
       "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": 28743.60748020786,
-      "required_pages_per_second_upper_estimate": 0.3686921342749305
+      "required_token_positions_per_second": 28737.98608865019,
+      "required_pages_per_second_upper_estimate": 0.36862002909980873
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06383001729673765,
-    "producer_p95_seconds": 0.21366545599812525,
-    "gpu_update_seconds": 0.9618834385710887,
-    "producer_batches_per_second": 15.666610199259807,
-    "gpu_batches_per_second": 1.0396270066626105,
-    "producer_headroom": 15.069452889216914,
+    "producer_batch_seconds": 0.0798577038125643,
+    "producer_p95_seconds": 0.22324003899848321,
+    "gpu_update_seconds": 0.9620715910541598,
+    "producer_batches_per_second": 12.522273397030313,
+    "gpu_batches_per_second": 1.0394236866554611,
+    "producer_headroom": 12.047323490696131,
     "sustainable": true,
-    "buffer_seconds": 60.59865662997859,
+    "buffer_seconds": 60.61051023641207,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454824.15206218616,
-    "upload_bytes_per_second": 5818588.066438202,
+    "generation_bytes_per_second": 453392.3475524882,
+    "upload_bytes_per_second": 5703426.921594854,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.59467526816547,
+    "estimated_full_shards_per_hour": 97.28744335108742,
     "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": 1.34454566200111,
-      "bytes": 16421464,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 0.9113021489993116,
       "bytes": 908789,
@@ -562,6 +556,12 @@
       "bytes": 1161196,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.0245182449980348,
+      "bytes": 6075842,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -167,3 +167,4 @@
 {"step": 413800, "loss": 0.5994492769241333, "lr": 0.0003, "grad_norm": 0.1538388363605086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60569432458, "tokens_per_second": 152868.82163962902, "data_wait_s": 1.0600000678095967e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 413900, "loss": 0.6144594550132751, "lr": 0.0003, "grad_norm": 0.20920066163767403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60584226589, "tokens_per_second": 153818.20431129623, "data_wait_s": 1.3491997378878295e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 414000, "loss": 0.5874106287956238, "lr": 0.0003, "grad_norm": 0.1685090484712039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60598881524, "tokens_per_second": 152361.95263786722, "data_wait_s": 1.8162998458137736e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 414100, "loss": 0.544400691986084, "lr": 0.0003, "grad_norm": 0.16984980498497781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60613406530, "tokens_per_second": 150581.64305898597, "data_wait_s": 1.7678998119663447e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:35b366f96d998dab5ee4ac5f3eb3c3f0e009e91f03f33fbaf5cb66d9a03df58d
-size 360456
+oid sha256:09bd8b7fdcf939b56713b4f535227d49d473e2ca77c7699a25c97620c459fba3
+size 362574