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": 76966.96224732761,
-      "required_pages_per_second_upper_estimate": 0.0053247530362271565
+      "required_token_positions_per_second": 76456.20887580977,
+      "required_pages_per_second_upper_estimate": 0.005289417932874454
     },
     {
       "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": 9620.870280915951,
-      "required_pages_per_second_upper_estimate": 0.004028057480002191
+      "required_token_positions_per_second": 9557.02610947622,
+      "required_pages_per_second_upper_estimate": 0.00400132725863828
     },
     {
       "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": 19241.740561831903,
-      "required_pages_per_second_upper_estimate": 0.004007547120207187
+      "required_token_positions_per_second": 19114.05221895244,
+      "required_pages_per_second_upper_estimate": 0.003980953005753443
     },
     {
       "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": 28862.610842747854,
-      "required_pages_per_second_upper_estimate": 0.0017302694493677507
+      "required_token_positions_per_second": 28671.078328428663,
+      "required_pages_per_second_upper_estimate": 0.001718787367587531
     },
     {
       "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": 28862.610842747854,
-      "required_pages_per_second_upper_estimate": 0.002243941039815377
+      "required_token_positions_per_second": 28671.078328428663,
+      "required_pages_per_second_upper_estimate": 0.0022290502292895563
     },
     {
       "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": 28862.610842747854,
-      "required_pages_per_second_upper_estimate": 0.37021858163373805
+      "required_token_positions_per_second": 28671.078328428663,
+      "required_pages_per_second_upper_estimate": 0.3677618081916428
     }
   ],
   "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": 0.36381145714844365,
-    "producer_p95_seconds": 0.28084187700005714,
-    "gpu_update_seconds": 0.9579174992392262,
-    "producer_batches_per_second": 2.748676492593185,
-    "gpu_batches_per_second": 1.0439312370785538,
-    "producer_headroom": 2.6330053120025116,
+    "producer_batch_seconds": 0.11135556822668491,
+    "producer_p95_seconds": 0.28215826700034086,
+    "gpu_update_seconds": 0.9643167125871847,
+    "producer_batches_per_second": 8.980242442518138,
+    "gpu_batches_per_second": 1.0370037011150413,
+    "producer_headroom": 8.659797870405,
     "sustainable": true,
-    "buffer_seconds": 60.34880245207125,
+    "buffer_seconds": 60.751952892992634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363697.46128506004,
-    "upload_bytes_per_second": 4832283.479201395,
+    "generation_bytes_per_second": 348612.423603408,
+    "upload_bytes_per_second": 4905498.855286173,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.0410087481866,
+    "estimated_full_shards_per_hour": 74.80411082340889,
     "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": 1.3772630089988525,
-      "bytes": 6008251,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.3044562169998244,
       "bytes": 6256652,
@@ -562,6 +556,12 @@
       "bytes": 1105976,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.2929117179992318,
+      "bytes": 11923823,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 786200, "loss": 0.5136899948120117, "lr": 3e-05, "grad_norm": 0.25140067303154884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115077383638, "tokens_per_second": 151663.62490066505, "data_wait_s": 1.0378000297350809e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 786300, "loss": 0.6177685856819153, "lr": 3e-05, "grad_norm": 0.2900455670641432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115091916283, "tokens_per_second": 151004.5800091677, "data_wait_s": 1.8485003238311037e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 786400, "loss": 0.5295373797416687, "lr": 3e-05, "grad_norm": 0.26525414383700247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115106625793, "tokens_per_second": 151712.22091129285, "data_wait_s": 1.5200999769149348e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 786500, "loss": 0.5489715337753296, "lr": 3e-05, "grad_norm": 0.256187664183011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115121285169, "tokens_per_second": 152298.42278352476, "data_wait_s": 1.4785000530537218e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ffe296e2c748ddf2089586ac0000fd3e8bd759b57d3b49e019e54d4425150d2
-size 563232
+oid sha256:97da669dfbcd12d7db52f93a339414600e762790032b8f4de2ff30335efed2cc
+size 565350