Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76563.57720234287,
-      "required_pages_per_second_upper_estimate": 0.005296845922832855
+      "required_token_positions_per_second": 76501.63186630252,
+      "required_pages_per_second_upper_estimate": 0.005292560400752601
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9570.447150292859,
-      "required_pages_per_second_upper_estimate": 0.0040069463681650045
+      "required_token_positions_per_second": 9562.703983287814,
+      "required_pages_per_second_upper_estimate": 0.004003704465835706
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19140.894300585718,
-      "required_pages_per_second_upper_estimate": 0.003986543503484359
+      "required_token_positions_per_second": 19125.40796657563,
+      "required_pages_per_second_upper_estimate": 0.003983318108512149
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28711.341450878575,
-      "required_pages_per_second_upper_estimate": 0.0017212010803001748
+      "required_token_positions_per_second": 28688.11194986344,
+      "required_pages_per_second_upper_estimate": 0.0017198085071837119
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28711.341450878575,
-      "required_pages_per_second_upper_estimate": 0.0022321805099612775
+      "required_token_positions_per_second": 28688.11194986344,
+      "required_pages_per_second_upper_estimate": 0.0022303745184331274
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28711.341450878575,
-      "required_pages_per_second_upper_estimate": 0.36827826029525756
+      "required_token_positions_per_second": 28688.11194986344,
+      "required_pages_per_second_upper_estimate": 0.36798029719813036
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08309998680431363,
-    "producer_p95_seconds": 0.24382342099852394,
-    "gpu_update_seconds": 0.9629644106772992,
-    "producer_batches_per_second": 12.03369625502866,
-    "gpu_batches_per_second": 1.038459977245319,
-    "producer_headroom": 11.588021222493296,
+    "producer_batch_seconds": 0.070331091343661,
+    "producer_p95_seconds": 0.2402992159986752,
+    "gpu_update_seconds": 0.9637441476915182,
+    "producer_batches_per_second": 14.218462715353994,
+    "gpu_batches_per_second": 1.0376197898532784,
+    "producer_headroom": 13.702960231092465,
     "sustainable": true,
-    "buffer_seconds": 60.66675787266985,
+    "buffer_seconds": 60.71588130456564,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326860.5565800503,
-    "upload_bytes_per_second": 5425391.340286881,
+    "generation_bytes_per_second": 340807.9157576396,
+    "upload_bytes_per_second": 5395658.049997949,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.13666651774533,
+    "estimated_full_shards_per_hour": 73.1294451193513,
     "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": 1.1401308700005757,
-      "bytes": 8309975,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2373918849989423,
       "bytes": 6097119,
@@ -562,6 +556,12 @@
       "bytes": 2751265,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.124440429997776,
+      "bytes": 5889635,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 581900, "loss": 0.5755289793014526, "lr": 0.0003, "grad_norm": 0.15386969171978265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85173342640, "tokens_per_second": 150449.20960974216, "data_wait_s": 9.972998668672517e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 582000, "loss": 0.56739342212677, "lr": 0.0003, "grad_norm": 0.163571416364174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85187903483, "tokens_per_second": 151341.70474212623, "data_wait_s": 1.1173000530106947e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 582100, "loss": 0.613437831401825, "lr": 0.0003, "grad_norm": 0.18855394660947497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85202590969, "tokens_per_second": 153180.8067054749, "data_wait_s": 1.205899752676487e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 582200, "loss": 0.6435660719871521, "lr": 0.0003, "grad_norm": 0.18179106690589025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85217344932, "tokens_per_second": 153247.3753550792, "data_wait_s": 7.557002390967682e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71b3b94da2aac3a9628979620f5fdf4fe84eb0bcc36efc52efbd6582c16dc554
-size 533580
+oid sha256:28a56ca7d6824aba516f7c2d021b334443690f6f7b38869b52bbc7e637289453
+size 535698