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": 77382.54942338195,
-      "required_pages_per_second_upper_estimate": 0.005353504321361698
+      "required_token_positions_per_second": 77526.4200042374,
+      "required_pages_per_second_upper_estimate": 0.005363457621971018
     },
     {
       "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": 9672.818677922744,
-      "required_pages_per_second_upper_estimate": 0.004049807189023047
+      "required_token_positions_per_second": 9690.802500529675,
+      "required_pages_per_second_upper_estimate": 0.004057336640003649
     },
     {
       "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": 19345.637355845487,
-      "required_pages_per_second_upper_estimate": 0.004029186082457505
+      "required_token_positions_per_second": 19381.60500105935,
+      "required_pages_per_second_upper_estimate": 0.004036677194424963
     },
     {
       "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": 29018.45603376823,
-      "required_pages_per_second_upper_estimate": 0.001739612130607594
+      "required_token_positions_per_second": 29072.407501589027,
+      "required_pages_per_second_upper_estimate": 0.0017428464387243294
     },
     {
       "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": 29018.45603376823,
-      "required_pages_per_second_upper_estimate": 0.002256057317926661
+      "required_token_positions_per_second": 29072.407501589027,
+      "required_pages_per_second_upper_estimate": 0.0022602518072423004
     },
     {
       "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": 29018.45603376823,
-      "required_pages_per_second_upper_estimate": 0.3722175964106185
+      "required_token_positions_per_second": 29072.407501589027,
+      "required_pages_per_second_upper_estimate": 0.3729096279112508
     }
   ],
   "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": 0.08185498214837139,
-    "producer_p95_seconds": 0.22784318099911616,
-    "gpu_update_seconds": 0.9527729513874392,
-    "producer_batches_per_second": 12.216727360435888,
-    "gpu_batches_per_second": 1.0495679989065476,
-    "producer_headroom": 11.639767383498182,
+    "producer_batch_seconds": 0.05954605658592982,
+    "producer_p95_seconds": 0.20479220299966983,
+    "gpu_update_seconds": 0.9510048315912203,
+    "producer_batches_per_second": 16.793723335094715,
+    "gpu_batches_per_second": 1.0515193685470567,
+    "producer_headroom": 15.970912032081296,
     "sustainable": true,
-    "buffer_seconds": 60.02469593740867,
+    "buffer_seconds": 60.8643092218381,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430368.1528417899,
-    "upload_bytes_per_second": 5329479.7713479,
+    "generation_bytes_per_second": 429742.01495864266,
+    "upload_bytes_per_second": 5300551.907985133,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.34698714199325,
+    "estimated_full_shards_per_hour": 92.2126325280138,
     "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": 1.1579751299996133,
-      "bytes": 6933490,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.4397287989995675,
       "bytes": 11874214,
@@ -562,6 +556,12 @@
       "bytes": 6746870,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.4409255560003658,
+      "bytes": 6062763,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -130,3 +130,4 @@
 {"step": 898000, "loss": 0.549811601638794, "lr": 3e-05, "grad_norm": 0.33915004162047574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131441966562, "tokens_per_second": 149895.1024518364, "data_wait_s": 1.588200029800646e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 898100, "loss": 0.5748760104179382, "lr": 3e-05, "grad_norm": 0.3712900950537548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131456510152, "tokens_per_second": 151927.77316547764, "data_wait_s": 1.4706000001751818e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 898200, "loss": 0.5086495876312256, "lr": 3e-05, "grad_norm": 0.3278603985220057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131470952327, "tokens_per_second": 151036.9326328292, "data_wait_s": 1.2135000361013226e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 898300, "loss": 0.559821367263794, "lr": 3e-05, "grad_norm": 0.3214628629964652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131485666654, "tokens_per_second": 153300.1605214211, "data_wait_s": 1.3654000213136896e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50de157690d1bab3bcc5d50bf2667b79c723b92aa207f95d3a6dbd43027b9a8f
-size 281307
+oid sha256:aeb333a3e4c355d4963fb28c18eec9b617fccd769d32536ee2a0c2079c02511d
+size 283425