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": 74759.99155725299,
-      "required_pages_per_second_upper_estimate": 0.006480006590716374
+      "required_token_positions_per_second": 74782.58916915893,
+      "required_pages_per_second_upper_estimate": 0.006481965294443258
     },
     {
       "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": 9344.998944656623,
-      "required_pages_per_second_upper_estimate": 0.003727859799208801
+      "required_token_positions_per_second": 9347.823646144867,
+      "required_pages_per_second_upper_estimate": 0.0037289866148655123
     },
     {
       "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": 18689.997889313247,
-      "required_pages_per_second_upper_estimate": 0.0043236020789642155
+      "required_token_positions_per_second": 18695.647292289734,
+      "required_pages_per_second_upper_estimate": 0.004324908968916729
     },
     {
       "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": 28034.99683396987,
-      "required_pages_per_second_upper_estimate": 0.0017071842189279572
+      "required_token_positions_per_second": 28043.470938434602,
+      "required_pages_per_second_upper_estimate": 0.0017077002474296691
     },
     {
       "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": 28034.99683396987,
-      "required_pages_per_second_upper_estimate": 0.002158211662117232
+      "required_token_positions_per_second": 28043.470938434602,
+      "required_pages_per_second_upper_estimate": 0.0021588640221367494
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28034.99683396987,
-      "required_pages_per_second_upper_estimate": 0.34810083357922283
+      "required_token_positions_per_second": 28043.470938434602,
+      "required_pages_per_second_upper_estimate": 0.3482060535956716
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10125474409375101,
-    "producer_p95_seconds": 0.3099013169994578,
-    "gpu_update_seconds": 0.986195938017694,
-    "producer_batches_per_second": 9.87608046368778,
-    "gpu_batches_per_second": 1.0139972813212483,
-    "producer_headroom": 9.739750436824792,
+    "producer_batch_seconds": 0.08681971811722633,
+    "producer_p95_seconds": 0.30607229299857863,
+    "gpu_update_seconds": 0.9858979318464964,
+    "producer_batches_per_second": 11.518120787374281,
+    "gpu_batches_per_second": 1.0143037810487052,
+    "producer_headroom": 11.355691463030443,
     "sustainable": true,
-    "buffer_seconds": 60.157952219079334,
+    "buffer_seconds": 60.13977384263628,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 198240.8278925489,
-    "upload_bytes_per_second": 6091143.954698812,
+    "generation_bytes_per_second": 221512.61115691747,
+    "upload_bytes_per_second": 5986981.186598207,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 42.53786685545302,
+    "estimated_full_shards_per_hour": 47.53144980459827,
     "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": 1.0367009409983439,
-      "bytes": 10231090,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.8111973979994218,
       "bytes": 875148,
@@ -562,6 +556,12 @@
       "bytes": 6097937,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0204584569983126,
+      "bytes": 2643615,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -183,3 +183,4 @@
 {"step": 18300, "loss": 0.7021921593695879, "lr": 0.0003, "grad_norm": 0.1532839326296825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2677681137, "tokens_per_second": 149770.87484752035, "data_wait_s": 1.052200241247192e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 18400, "loss": 0.6478817276656628, "lr": 0.0003, "grad_norm": 0.14570998880915342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2692390728, "tokens_per_second": 149680.40948064916, "data_wait_s": 1.0278999980073422e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 18500, "loss": 0.7289327643811703, "lr": 0.0003, "grad_norm": 0.1548255705400022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2707145406, "tokens_per_second": 149209.71046541084, "data_wait_s": 9.69899701885879e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 18600, "loss": 0.6644454542547464, "lr": 0.0003, "grad_norm": 0.15401647829993756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2721860495, "tokens_per_second": 149481.10288003285, "data_wait_s": 1.3980999938212335e-05, "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:a55e8825d91aa619c3f5d332623f560383d3f81f7c6295a346ee7fafe2e20745
-size 388570
+oid sha256:6b044981a431a8cccb493248acf7970f4adffe66c31d0f8eebcbf9d5a2fe90df
+size 390688