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": 74389.06575289226,
-      "required_pages_per_second_upper_estimate": 0.006447855682097243
+      "required_token_positions_per_second": 74547.30306075996,
+      "required_pages_per_second_upper_estimate": 0.006461571290894425
     },
     {
       "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": 9298.633219111533,
-      "required_pages_per_second_upper_estimate": 0.003709363818059491
+      "required_token_positions_per_second": 9318.412882594996,
+      "required_pages_per_second_upper_estimate": 0.0037172542215553675
     },
     {
       "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": 18597.266438223065,
-      "required_pages_per_second_upper_estimate": 0.004302150289772301
+      "required_token_positions_per_second": 18636.82576518999,
+      "required_pages_per_second_upper_estimate": 0.004311301643845723
     },
     {
       "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": 27895.8996573346,
-      "required_pages_per_second_upper_estimate": 0.001698713930657354
+      "required_token_positions_per_second": 27955.238647784987,
+      "required_pages_per_second_upper_estimate": 0.0017023273638481604
     },
     {
       "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": 27895.8996573346,
-      "required_pages_per_second_upper_estimate": 0.0021475035764142196
+      "required_token_positions_per_second": 27955.238647784987,
+      "required_pages_per_second_upper_estimate": 0.002152071656159929
     },
     {
       "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": 27895.8996573346,
-      "required_pages_per_second_upper_estimate": 0.34637371217371643
+      "required_token_positions_per_second": 27955.238647784987,
+      "required_pages_per_second_upper_estimate": 0.3471105038402844
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.22863940589857634,
-    "producer_p95_seconds": 0.2969011440000031,
-    "gpu_update_seconds": 0.991113401597377,
-    "producer_batches_per_second": 4.373699258314188,
-    "gpu_batches_per_second": 1.0089662781154007,
-    "producer_headroom": 4.3348319494717,
+    "producer_batch_seconds": 0.2834814608830527,
+    "producer_p95_seconds": 0.30807454800014966,
+    "gpu_update_seconds": 0.989009621715058,
+    "producer_batches_per_second": 3.5275675414010212,
+    "gpu_batches_per_second": 1.0111125089621307,
+    "producer_headroom": 3.4887982396953414,
     "sustainable": true,
-    "buffer_seconds": 60.45791749744,
+    "buffer_seconds": 60.32958692461854,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 167,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 574167.803898092,
-    "upload_bytes_per_second": 5722526.054790809,
+    "generation_bytes_per_second": 558506.229155328,
+    "upload_bytes_per_second": 5658236.221536857,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.60152238706145,
+    "estimated_full_shards_per_hour": 119.84243541712647,
     "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": 1.3119218800002272,
-      "bytes": 12680620,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.8689146019987675,
       "bytes": 1090693,
@@ -562,6 +556,12 @@
       "bytes": 6024431,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0379702050013293,
+      "bytes": 6386142,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -171,3 +171,4 @@
 {"step": 372100, "loss": 0.6335708033293486, "lr": 0.0003, "grad_norm": 0.18131454470469277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54464514533, "tokens_per_second": 148717.74400200957, "data_wait_s": 9.993000276153907e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 372200, "loss": 0.5570940934121609, "lr": 0.0003, "grad_norm": 0.14605080317111024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54479184826, "tokens_per_second": 148065.656737445, "data_wait_s": 7.60500188334845e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 372300, "loss": 0.5817138999700546, "lr": 0.0003, "grad_norm": 0.17083567174265302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54493951271, "tokens_per_second": 150660.22467692784, "data_wait_s": 1.0944000678136945e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 372400, "loss": 0.5703586395829916, "lr": 0.0003, "grad_norm": 0.1545383032913337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54508489145, "tokens_per_second": 148005.0871590917, "data_wait_s": 1.5253001038217917e-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:aecabd9d1f5b2e13d010c66de37968be0a124bd532adee21a76f9264f9f3dd90
-size 368928
+oid sha256:e27737f041a66edde448357d9d09adcbadb5e4c4a77cc5f94bb557aca7cbd7c3
+size 371046