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": 77560.71680571201,
-      "required_pages_per_second_upper_estimate": 0.005365830354276575
+      "required_token_positions_per_second": 75971.16134754471,
+      "required_pages_per_second_upper_estimate": 0.005255861219403753
     },
     {
       "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": 9695.089600714002,
-      "required_pages_per_second_upper_estimate": 0.004059131559325994
+      "required_token_positions_per_second": 9496.395168443089,
+      "required_pages_per_second_upper_estimate": 0.003975942349745731
     },
     {
       "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": 19390.179201428004,
-      "required_pages_per_second_upper_estimate": 0.004038462974244884
+      "required_token_positions_per_second": 18992.790336886177,
+      "required_pages_per_second_upper_estimate": 0.0039556973536099185
     },
     {
       "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": 29085.268802142007,
-      "required_pages_per_second_upper_estimate": 0.0017436174540544115
+      "required_token_positions_per_second": 28489.185505329267,
+      "required_pages_per_second_upper_estimate": 0.0017078831705769817
     },
     {
       "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": 29085.268802142007,
-      "required_pages_per_second_upper_estimate": 0.002261251716789412
+      "required_token_positions_per_second": 28489.185505329267,
+      "required_pages_per_second_upper_estimate": 0.002214908862355554
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 29085.268802142007,
-      "required_pages_per_second_upper_estimate": 0.3730745988653558
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28489.185505329267,
+      "required_pages_per_second_upper_estimate": 0.365428682358221
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0952288898830318,
-    "producer_p95_seconds": 0.22767995799949858,
-    "gpu_update_seconds": 0.9505843039677818,
-    "producer_batches_per_second": 10.501014988500705,
-    "gpu_batches_per_second": 1.0519845486885853,
-    "producer_headroom": 9.982100023799186,
+    "producer_batch_seconds": 0.4784675539687271,
+    "producer_p95_seconds": 0.23298569599865004,
+    "gpu_update_seconds": 0.9704735151107808,
+    "producer_batches_per_second": 2.090005877525732,
+    "gpu_batches_per_second": 1.0304248229647448,
+    "producer_headroom": 2.0282953505645893,
     "sustainable": true,
-    "buffer_seconds": 60.837395453938036,
+    "buffer_seconds": 60.16935793686841,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280142.2736768203,
-    "upload_bytes_per_second": 5036609.968266217,
+    "generation_bytes_per_second": 237083.44740828767,
+    "upload_bytes_per_second": 5126031.569571022,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.11201055267769,
+    "estimated_full_shards_per_hour": 50.87258879362557,
     "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": 1.144102094000118,
-      "bytes": 5891615,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.2515704560009908,
       "bytes": 7512167,
@@ -562,6 +556,12 @@
       "bytes": 2438669,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.6140179690009973,
+      "bytes": 15868305,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -178,3 +178,4 @@
 {"step": 902800, "loss": 0.5254056453704834, "lr": 3e-05, "grad_norm": 0.31550267280815525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132143983708, "tokens_per_second": 151150.56165563347, "data_wait_s": 2.009699892369099e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 902900, "loss": 0.5308130979537964, "lr": 3e-05, "grad_norm": 0.3173862850624366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132158562799, "tokens_per_second": 152321.57899044192, "data_wait_s": 1.182200139737688e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 903000, "loss": 0.5116169452667236, "lr": 3e-05, "grad_norm": 0.32104813072337285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132173108896, "tokens_per_second": 113058.53595252754, "data_wait_s": 1.0078001650981605e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 903100, "loss": 0.5306736826896667, "lr": 3e-05, "grad_norm": 0.31442032819851684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132187761341, "tokens_per_second": 152746.940810981, "data_wait_s": 1.054000313160941e-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:ec2ca4c01eda2e2270d205c7a3d44a7fad0ea906ba525711feb3f0912c80df17
-size 383754
+oid sha256:120735a05f244ba727ea8e401e90af3b6c148e057f8f9381a558383d08e06b9a
+size 385872