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": 77455.234641437,
-      "required_pages_per_second_upper_estimate": 0.005358532853399688
+      "required_token_positions_per_second": 77339.46595289434,
+      "required_pages_per_second_upper_estimate": 0.005350523706905941
     },
     {
       "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": 9681.904330179625,
-      "required_pages_per_second_upper_estimate": 0.004053611161893009
+      "required_token_positions_per_second": 9667.433244111793,
+      "required_pages_per_second_upper_estimate": 0.004047552420347048
     },
     {
       "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": 19363.80866035925,
-      "required_pages_per_second_upper_estimate": 0.004032970685978196
+      "required_token_positions_per_second": 19334.866488223586,
+      "required_pages_per_second_upper_estimate": 0.004026942794778744
     },
     {
       "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": 29045.712990538872,
-      "required_pages_per_second_upper_estimate": 0.0017412461435470307
+      "required_token_positions_per_second": 29002.29973233538,
+      "required_pages_per_second_upper_estimate": 0.0017386435850059387
     },
     {
       "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": 29045.712990538872,
-      "required_pages_per_second_upper_estimate": 0.002258176426424896
+      "required_token_positions_per_second": 29002.29973233538,
+      "required_pages_per_second_upper_estimate": 0.002254801236554317
     },
     {
       "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": 29045.712990538872,
-      "required_pages_per_second_upper_estimate": 0.3725672193858323
+      "required_token_positions_per_second": 29002.29973233538,
+      "required_pages_per_second_upper_estimate": 0.3720103607231229
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11044215324983497,
-    "producer_p95_seconds": 0.2851377519982634,
-    "gpu_update_seconds": 0.9518788541705224,
-    "producer_batches_per_second": 9.054513793640602,
-    "gpu_batches_per_second": 1.0505538552712266,
-    "producer_headroom": 8.618800214961807,
+    "producer_batch_seconds": 0.13084672797634767,
+    "producer_p95_seconds": 0.2919916480022948,
+    "gpu_update_seconds": 0.9533037122974962,
+    "producer_batches_per_second": 7.64252966402617,
+    "gpu_batches_per_second": 1.0489836419392138,
+    "producer_headroom": 7.285651900059885,
     "sustainable": true,
-    "buffer_seconds": 60.920246666913435,
+    "buffer_seconds": 60.05813387474226,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334745.1260276774,
-    "upload_bytes_per_second": 3911853.633855266,
+    "generation_bytes_per_second": 385573.21953660017,
+    "upload_bytes_per_second": 3888200.061676735,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.82851157782308,
+    "estimated_full_shards_per_hour": 82.73503722737793,
     "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": 1.308492041000136,
-      "bytes": 6108009,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 0.922122173000389,
       "bytes": 1095644,
@@ -562,6 +556,12 @@
       "bytes": 6018270,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0365287800013903,
+      "bytes": 2455252,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 555900, "loss": 0.5112012624740601, "lr": 0.0003, "grad_norm": 0.14630499997509794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81365821327, "tokens_per_second": 153208.73713184643, "data_wait_s": 1.0210998880211264e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 556000, "loss": 0.6268565058708191, "lr": 0.0003, "grad_norm": 0.17406294932042216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81380463327, "tokens_per_second": 152907.1800330645, "data_wait_s": 1.559000156703405e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 556100, "loss": 0.5874567031860352, "lr": 0.0003, "grad_norm": 0.15580764648370396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81395148084, "tokens_per_second": 152946.02463316132, "data_wait_s": 1.6943999071372673e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 556200, "loss": 0.5754328966140747, "lr": 0.0003, "grad_norm": 0.1559919298696575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81409812675, "tokens_per_second": 151519.1952331492, "data_wait_s": 9.864001185633242e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e771eefdd822f584cdda3f01ac0c15118bb81977d6ef6adb249e4f59b6b1456e
-size 663561
+oid sha256:170f474dba8b1f042368dc9e9dbb481c5890e694bef925ed07941e0f37340317
+size 665679