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": 76108.58541228583,
-      "required_pages_per_second_upper_estimate": 0.006596899288090744
+      "required_token_positions_per_second": 76258.54207205445,
+      "required_pages_per_second_upper_estimate": 0.0066098971512978085
     },
     {
       "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": 9513.573176535729,
-      "required_pages_per_second_upper_estimate": 0.0037951065807147475
+      "required_token_positions_per_second": 9532.317759006806,
+      "required_pages_per_second_upper_estimate": 0.003802584074918943
     },
     {
       "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": 19027.146353071457,
-      "required_pages_per_second_upper_estimate": 0.004401595442444373
+      "required_token_positions_per_second": 19064.635518013612,
+      "required_pages_per_second_upper_estimate": 0.004410267900966973
     },
     {
       "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": 28540.719529607188,
-      "required_pages_per_second_upper_estimate": 0.0017379800777703451
+      "required_token_positions_per_second": 28596.95327702042,
+      "required_pages_per_second_upper_estimate": 0.00174140441795214
     },
     {
       "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": 28540.719529607188,
-      "required_pages_per_second_upper_estimate": 0.0021971435951574133
+      "required_token_positions_per_second": 28596.95327702042,
+      "required_pages_per_second_upper_estimate": 0.00220147262469826
     },
     {
       "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": 28540.719529607188,
-      "required_pages_per_second_upper_estimate": 0.35438021691405425
+      "required_token_positions_per_second": 28596.95327702042,
+      "required_pages_per_second_upper_estimate": 0.35507845185468073
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,17 +155,17 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.05422725875000012,
-    "producer_p95_seconds": 0.14343000600001687,
-    "gpu_update_seconds": 0.9687211974918464,
-    "producer_batches_per_second": 18.44091003401491,
-    "gpu_batches_per_second": 1.032288756134519,
-    "producer_headroom": 17.86410045099033,
+    "producer_batch_seconds": 0.048913495023438,
+    "producer_p95_seconds": 0.12456286900032865,
+    "gpu_update_seconds": 0.9668162804677879,
+    "producer_batches_per_second": 20.444255711451973,
+    "gpu_batches_per_second": 1.0343226735033426,
+    "producer_headroom": 19.765839263878323,
     "sustainable": true,
-    "buffer_seconds": 60.06071424449448,
+    "buffer_seconds": 60.909425669470636,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 68,

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -15,3 +15,4 @@
 {"step": 356500, "loss": 0.6198965087532997, "lr": 0.0003, "grad_norm": 0.17267438068680266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52181420610, "tokens_per_second": 149105.94934928452, "data_wait_s": 1.1697999980242457e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356600, "loss": 0.6039838008582592, "lr": 0.0003, "grad_norm": 0.15690151570082372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52196046402, "tokens_per_second": 150427.10282492568, "data_wait_s": 1.3294999916979577e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356700, "loss": 0.6328282952308655, "lr": 0.0003, "grad_norm": 0.19027880083121018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52210646056, "tokens_per_second": 150503.8607225417, "data_wait_s": 7.793999884597724e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 356800, "loss": 0.5971382074058056, "lr": 0.0003, "grad_norm": 0.17656761639722304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52225208874, "tokens_per_second": 149479.63006219224, "data_wait_s": 1.575900023453869e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebcfefbaccac99076050d8b6e7790b6fadd09af062d98fb225dd1a145d770c81
-size 36171
+oid sha256:623f844b99a2a16861c70aa43ef1882d3a06cdcc55e9dc27be9611d026e63bfa
+size 38289