Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75768.99073357528,
-      "required_pages_per_second_upper_estimate": 0.006567464082034958
+      "required_token_positions_per_second": 75403.0773433573,
+      "required_pages_per_second_upper_estimate": 0.006535747636769873
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9471.12384169691,
-      "required_pages_per_second_upper_estimate": 0.0037781729063734283
+      "required_token_positions_per_second": 9425.384667919663,
+      "required_pages_per_second_upper_estimate": 0.0037599268660920946
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18942.24768339382,
-      "required_pages_per_second_upper_estimate": 0.004381955629380006
+      "required_token_positions_per_second": 18850.769335839326,
+      "required_pages_per_second_upper_estimate": 0.004360793723637204
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28413.371525090726,
-      "required_pages_per_second_upper_estimate": 0.0017302252524386355
+      "required_token_positions_per_second": 28276.154003758988,
+      "required_pages_per_second_upper_estimate": 0.0017218694253142283
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28413.371525090726,
-      "required_pages_per_second_upper_estimate": 0.0021873399932478975
+      "required_token_positions_per_second": 28276.154003758988,
+      "required_pages_per_second_upper_estimate": 0.00217677660861337
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28413.371525090726,
-      "required_pages_per_second_upper_estimate": 0.3527989809043139
+      "required_token_positions_per_second": 28276.154003758988,
+      "required_pages_per_second_upper_estimate": 0.3510951985268757
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1253010482500656,
-    "producer_p95_seconds": 0.3474264549986401,
-    "gpu_update_seconds": 0.9730629811244027,
-    "producer_batches_per_second": 7.980779203093989,
-    "gpu_batches_per_second": 1.0276827085174598,
-    "producer_headroom": 7.765800803058272,
+    "producer_batch_seconds": 0.15604042967981968,
+    "producer_p95_seconds": 0.3191753229984897,
+    "gpu_update_seconds": 0.9777850267870417,
+    "producer_batches_per_second": 6.408595529068371,
+    "gpu_batches_per_second": 1.0227196905294773,
+    "producer_headroom": 6.266228751057433,
     "sustainable": true,
-    "buffer_seconds": 60.329904829712966,
+    "buffer_seconds": 60.622671660796584,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440363.4417499805,
-    "upload_bytes_per_second": 5727427.328722881,
+    "generation_bytes_per_second": 447853.8335587134,
+    "upload_bytes_per_second": 5815124.3972679665,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.49174346327364,
+    "estimated_full_shards_per_hour": 96.0990071780305,
     "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": 1.0843618420003622,
-      "bytes": 5928833,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.330314847000409,
       "bytes": 13379246,
@@ -562,6 +556,12 @@
       "bytes": 2356425,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0524133049984812,
+      "bytes": 12178249,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 24500, "loss": 0.6439226306974888, "lr": 0.0003, "grad_norm": 0.13887686574509203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3585136027, "tokens_per_second": 147514.78263566113, "data_wait_s": 8.819999493425712e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 24600, "loss": 0.6646617352962494, "lr": 0.0003, "grad_norm": 0.13959699323131255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3599512906, "tokens_per_second": 146573.92568271855, "data_wait_s": 8.941999112721533e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 24700, "loss": 0.6467013955116272, "lr": 0.0003, "grad_norm": 0.13487899901526593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3614016837, "tokens_per_second": 146990.5511808094, "data_wait_s": 1.1520001862663776e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 24800, "loss": 0.6468101106584072, "lr": 0.0003, "grad_norm": 0.13603707268163948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3628643746, "tokens_per_second": 148611.72245921195, "data_wait_s": 9.395000233780593e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cddadfcd14e9336a960c98b4a7a164d6102fbfc631aef4e6ab2b97478c04cd73
-size 520669
+oid sha256:07c528458550a22e4019194797cff3c78852735ea3a7f8e112195e2a9ce114b5
+size 522787