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": 76180.0364056722,
-      "required_pages_per_second_upper_estimate": 0.00527031168058192
+      "required_token_positions_per_second": 76075.3752010543,
+      "required_pages_per_second_upper_estimate": 0.005263070975599001
     },
     {
       "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": 9522.504550709025,
-      "required_pages_per_second_upper_estimate": 0.003986873802874578
+      "required_token_positions_per_second": 9509.421900131787,
+      "required_pages_per_second_upper_estimate": 0.003981396370274698
     },
     {
       "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": 19045.00910141805,
-      "required_pages_per_second_upper_estimate": 0.003966573145160481
+      "required_token_positions_per_second": 19018.843800263574,
+      "required_pages_per_second_upper_estimate": 0.003961123602955397
     },
     {
       "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": 28567.51365212708,
-      "required_pages_per_second_upper_estimate": 0.0017125788233773554
+      "required_token_positions_per_second": 28528.26570039536,
+      "required_pages_per_second_upper_estimate": 0.0017102259685991912
     },
     {
       "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": 28567.51365212708,
-      "required_pages_per_second_upper_estimate": 0.002220998531239977
+      "required_token_positions_per_second": 28528.26570039536,
+      "required_pages_per_second_upper_estimate": 0.0022179471756263312
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28567.51365212708,
-      "required_pages_per_second_upper_estimate": 0.36643339172313183
+      "required_token_positions_per_second": 28528.26570039536,
+      "required_pages_per_second_upper_estimate": 0.3659299611394846
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10068165421861863,
-    "producer_p95_seconds": 0.2587879359998624,
-    "gpu_update_seconds": 0.9678126117895943,
-    "producer_batches_per_second": 9.932296084732727,
-    "gpu_batches_per_second": 1.0332578722557537,
-    "producer_headroom": 9.612601414832742,
+    "producer_batch_seconds": 0.08939356210163396,
+    "producer_p95_seconds": 0.2542347839989816,
+    "gpu_update_seconds": 0.9691440864425501,
+    "producer_batches_per_second": 11.186487891186983,
+    "gpu_batches_per_second": 1.0318383138163831,
+    "producer_headroom": 10.841318587805059,
     "sustainable": true,
-    "buffer_seconds": 60.00438193095485,
+    "buffer_seconds": 60.086933359438106,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367109.47845403635,
-    "upload_bytes_per_second": 4921221.752298457,
+    "generation_bytes_per_second": 367448.05649861845,
+    "upload_bytes_per_second": 4917165.025646714,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.77314820495431,
+    "estimated_full_shards_per_hour": 78.84579917162814,
     "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": 0.7830869729987171,
-      "bytes": 1161196,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.0245182449980348,
       "bytes": 6075842,
@@ -562,6 +556,12 @@
       "bytes": 10288628,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.1022804700005508,
+      "bytes": 2387118,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -244,3 +244,4 @@
 {"step": 421500, "loss": 0.5923106670379639, "lr": 0.0003, "grad_norm": 0.15291946229540204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61696290298, "tokens_per_second": 150591.54407922612, "data_wait_s": 7.5549978646449745e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 421600, "loss": 0.5581619739532471, "lr": 0.0003, "grad_norm": 0.15593133275044313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61710802049, "tokens_per_second": 149947.38246147762, "data_wait_s": 8.893999620340765e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 421700, "loss": 0.5913926959037781, "lr": 0.0003, "grad_norm": 0.17915203021932297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61725345966, "tokens_per_second": 151287.51366720232, "data_wait_s": 8.082999556791037e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 421800, "loss": 0.6039777398109436, "lr": 0.0003, "grad_norm": 0.1980378037377272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61740025221, "tokens_per_second": 151331.86318145675, "data_wait_s": 1.2757001968566328e-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 sha256:48a5ec8b27ca4d4d529fc8a27696b4408d89491e159be5a0ab85e090d1516e62
-size 525108
+oid sha256:f75e4c6587bf3cf01677488bd1621a705afa6a986a43decb3edcc94238d994ed
+size 527226