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": 75825.45888822727,
-      "required_pages_per_second_upper_estimate": 0.006572358598562963
+      "required_token_positions_per_second": 75627.12063873814,
+      "required_pages_per_second_upper_estimate": 0.006555167141780943
     },
     {
       "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": 9478.182361028408,
-      "required_pages_per_second_upper_estimate": 0.0037809886552690316
+      "required_token_positions_per_second": 9453.390079842267,
+      "required_pages_per_second_upper_estimate": 0.0037710986436262435
     },
     {
       "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": 18956.364722056816,
-      "required_pages_per_second_upper_estimate": 0.004385221357823295
+      "required_token_positions_per_second": 18906.780159684535,
+      "required_pages_per_second_upper_estimate": 0.004373750841976961
     },
     {
       "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": 28434.547083085225,
-      "required_pages_per_second_upper_estimate": 0.001731514732820934
+      "required_token_positions_per_second": 28360.1702395268,
+      "required_pages_per_second_upper_estimate": 0.0017269855732733657
     },
     {
       "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": 28434.547083085225,
-      "required_pages_per_second_upper_estimate": 0.00218897014632001
+      "required_token_positions_per_second": 28360.1702395268,
+      "required_pages_per_second_upper_estimate": 0.002183244411014602
     },
     {
       "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": 28434.547083085225,
-      "required_pages_per_second_upper_estimate": 0.35306191046457186
+      "required_token_positions_per_second": 28360.1702395268,
+      "required_pages_per_second_upper_estimate": 0.35213839899085886
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.08876055239838365,
-    "producer_p95_seconds": 0.28822078299708664,
-    "gpu_update_seconds": 0.9723383291182044,
-    "producer_batches_per_second": 11.26626607179847,
-    "gpu_batches_per_second": 1.0284486068824228,
-    "producer_headroom": 10.95462232765364,
+    "producer_batch_seconds": 0.07432529642954933,
+    "producer_p95_seconds": 0.2999383440001111,
+    "gpu_update_seconds": 0.9748883651433722,
+    "producer_batches_per_second": 13.454369481699537,
+    "gpu_batches_per_second": 1.025758472205107,
+    "producer_headroom": 13.11650826804894,
     "sustainable": true,
-    "buffer_seconds": 60.28497640532867,
+    "buffer_seconds": 60.44307863888908,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311538.20929310325,
-    "upload_bytes_per_second": 5597784.964530512,
+    "generation_bytes_per_second": 310727.0559268336,
+    "upload_bytes_per_second": 5601358.603867205,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.84884747595619,
+    "estimated_full_shards_per_hour": 66.67479284623867,
     "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": 1.1301041130009253,
-      "bytes": 5919931,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.7547573850006302,
       "bytes": 662301,
@@ -562,6 +556,12 @@
       "bytes": 12042987,
       "pages": 27,
       "shards": 4
+    },
+    {
+      "seconds": 1.1372878729998774,
+      "bytes": 6231473,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -228,3 +228,4 @@
 {"step": 337800, "loss": 0.5727712158113718, "lr": 0.0003, "grad_norm": 0.13993599349491304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49445122785, "tokens_per_second": 120803.95662036254, "data_wait_s": 1.0658997780410573e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 337900, "loss": 0.5599334165453911, "lr": 0.0003, "grad_norm": 0.1568640515650128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49459646170, "tokens_per_second": 147810.92229261424, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 338000, "loss": 0.5721419248729944, "lr": 0.0003, "grad_norm": 0.14536312385044528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49474213568, "tokens_per_second": 148280.6151080045, "data_wait_s": 1.296499976888299e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 338100, "loss": 0.5749847572296858, "lr": 0.0003, "grad_norm": 0.16412448464579407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49488941651, "tokens_per_second": 148960.1400757357, "data_wait_s": 1.1770996934501454e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c3660a3ef4e74559ffac3e70faa38897c6d4e3d12f6730220c5fd798250f43b5
-size 490437
+oid sha256:8f974f3ebfc7aeaefbb92e546a6ac75fac357520af9169ea342e9d9726303ca0
+size 492555