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": 75479.39113630546,
-      "required_pages_per_second_upper_estimate": 0.0065423623229270795
+      "required_token_positions_per_second": 75030.69939996467,
+      "required_pages_per_second_upper_estimate": 0.006503470860419868
     },
     {
       "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": 9434.923892038183,
-      "required_pages_per_second_upper_estimate": 0.0037637322052170826
+      "required_token_positions_per_second": 9378.837424995583,
+      "required_pages_per_second_upper_estimate": 0.0037413584749463793
     },
     {
       "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": 18869.847784076366,
-      "required_pages_per_second_upper_estimate": 0.004365207186867622
+      "required_token_positions_per_second": 18757.674849991166,
+      "required_pages_per_second_upper_estimate": 0.004339257952743225
     },
     {
       "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": 28304.77167611455,
-      "required_pages_per_second_upper_estimate": 0.001723612091415359
+      "required_token_positions_per_second": 28136.512274986748,
+      "required_pages_per_second_upper_estimate": 0.0017133659766755281
     },
     {
       "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": 28304.77167611455,
-      "required_pages_per_second_upper_estimate": 0.0021789796762500886
+      "required_token_positions_per_second": 28136.512274986748,
+      "required_pages_per_second_upper_estimate": 0.002166026601779437
     },
     {
       "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": 28304.77167611455,
-      "required_pages_per_second_upper_estimate": 0.35145053424034356
+      "required_token_positions_per_second": 28136.512274986748,
+      "required_pages_per_second_upper_estimate": 0.3493613156063269
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.10161335564083629,
-    "producer_p95_seconds": 0.40145624699653126,
-    "gpu_update_seconds": 0.9767964326428827,
-    "producer_batches_per_second": 9.841226024801417,
-    "gpu_batches_per_second": 1.0237547625909487,
-    "producer_headroom": 9.612874473858323,
+    "producer_batch_seconds": 0.1539512535004519,
+    "producer_p95_seconds": 0.37864449800690636,
+    "gpu_update_seconds": 0.9826377814630197,
+    "producer_batches_per_second": 6.495562571025539,
+    "gpu_batches_per_second": 1.0176689914274721,
+    "producer_headroom": 6.382785194146765,
     "sustainable": true,
-    "buffer_seconds": 60.56137882385873,
+    "buffer_seconds": 60.923542450707224,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430375.8343578834,
-    "upload_bytes_per_second": 5723475.89742017,
+    "generation_bytes_per_second": 441061.73271859775,
+    "upload_bytes_per_second": 5672077.061435079,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.34863541653039,
+    "estimated_full_shards_per_hour": 94.64158044975709,
     "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": 1.3006846259995655,
-      "bytes": 12006518,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.8710538570012432,
       "bytes": 5900566,
@@ -562,6 +556,12 @@
       "bytes": 11954791,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1746257199993124,
+      "bytes": 7528355,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -391,3 +391,4 @@
 {"step": 394100, "loss": 0.5695737525820732, "lr": 0.0003, "grad_norm": 0.14450641393979294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57684828750, "tokens_per_second": 149619.47986691375, "data_wait_s": 1.612699998077005e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 394200, "loss": 0.5794911608099937, "lr": 0.0003, "grad_norm": 0.14452503664774535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57699487255, "tokens_per_second": 149386.44426603787, "data_wait_s": 1.0415998985990882e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 394300, "loss": 0.5949999503791332, "lr": 0.0003, "grad_norm": 0.1640105266501889, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57714118574, "tokens_per_second": 149501.82989230246, "data_wait_s": 2.463500277372077e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 394400, "loss": 0.5831971168518066, "lr": 0.0003, "grad_norm": 0.14931650550310513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57728767715, "tokens_per_second": 149608.82075279692, "data_wait_s": 9.424999007023871e-06, "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:a962ec616579f140fe53b08abe0419051ddeb267291f8acdcf103c0fd50c3183
-size 838020
+oid sha256:1e6552046574064de2e7af8c76bb24f3a76ee59b799bfeea5d698b285cc30c3c
+size 840138