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": 77349.15431738223,
-      "required_pages_per_second_upper_estimate": 0.005351193970441316
+      "required_token_positions_per_second": 77440.02101516243,
+      "required_pages_per_second_upper_estimate": 0.005357480339433537
     },
     {
       "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": 9668.644289672779,
-      "required_pages_per_second_upper_estimate": 0.004048059459833926
+      "required_token_positions_per_second": 9680.002626895304,
+      "required_pages_per_second_upper_estimate": 0.004052814958439945
     },
     {
       "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": 19337.288579345557,
-      "required_pages_per_second_upper_estimate": 0.004027447252484612
+      "required_token_positions_per_second": 19360.00525379061,
+      "required_pages_per_second_upper_estimate": 0.004032178536692525
     },
     {
       "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": 29005.932869018336,
-      "required_pages_per_second_upper_estimate": 0.0017388613860026031
+      "required_token_positions_per_second": 29040.007880685913,
+      "required_pages_per_second_upper_estimate": 0.0017409041309225442
     },
     {
       "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": 29005.932869018336,
-      "required_pages_per_second_upper_estimate": 0.002255083696950937
+      "required_token_positions_per_second": 29040.007880685913,
+      "required_pages_per_second_upper_estimate": 0.0022577328792279545
     },
     {
       "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": 29005.932869018336,
-      "required_pages_per_second_upper_estimate": 0.37205696269953353
+      "required_token_positions_per_second": 29040.007880685913,
+      "required_pages_per_second_upper_estimate": 0.37249404036230827
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12325689775769888,
-    "producer_p95_seconds": 0.2230575850007881,
-    "gpu_update_seconds": 0.9531843062882916,
-    "producer_batches_per_second": 8.113136207320599,
-    "gpu_batches_per_second": 1.0491150487926193,
-    "producer_headroom": 7.733314107597307,
+    "producer_batch_seconds": 0.06017359400789246,
+    "producer_p95_seconds": 0.2042716260002635,
+    "gpu_update_seconds": 0.9520658573370527,
+    "producer_batches_per_second": 16.61858521976996,
+    "gpu_batches_per_second": 1.050347507258605,
+    "producer_headroom": 15.821987584989161,
     "sustainable": true,
-    "buffer_seconds": 60.05061129616237,
+    "buffer_seconds": 60.932214869571375,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387669.3345364263,
-    "upload_bytes_per_second": 3703354.0885011475,
+    "generation_bytes_per_second": 373440.9908655677,
+    "upload_bytes_per_second": 3650933.9170539975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.18481471128074,
+    "estimated_full_shards_per_hour": 80.13174337840341,
     "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": 1.1856751459999941,
-      "bytes": 11039128,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.5634893169990391,
       "bytes": 7514457,
@@ -562,6 +556,12 @@
       "bytes": 6391888,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.6235010039999906,
+      "bytes": 6476306,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -174,3 +174,4 @@
 {"step": 942400, "loss": 0.520447313785553, "lr": 3e-05, "grad_norm": 0.3352271217179194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137940681345, "tokens_per_second": 146237.1304412144, "data_wait_s": 1.642300048843026e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 942500, "loss": 0.5312053561210632, "lr": 3e-05, "grad_norm": 0.3516962228627744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137955413024, "tokens_per_second": 153717.4198919749, "data_wait_s": 1.5871002688072622e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 942600, "loss": 0.47934281826019287, "lr": 3e-05, "grad_norm": 0.3301637447809434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137970109623, "tokens_per_second": 153375.35597927953, "data_wait_s": 1.4600998838432133e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 942700, "loss": 0.5388817191123962, "lr": 3e-05, "grad_norm": 0.35459493587676943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137984743582, "tokens_per_second": 152744.6003907176, "data_wait_s": 1.1482999980216846e-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 sha256:7fbf87ac19ac0910efdad132252f4a21971e3207b41d964a6bc2d374c927abee
-size 375282
+oid sha256:7ff088cbabe4b97324ba3e7a00986a97118f6933dfd1445ab05158a904209565
+size 377400