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": 74961.74951749692,
-      "required_pages_per_second_upper_estimate": 0.006497494459359443
+      "required_token_positions_per_second": 75518.75880574604,
+      "required_pages_per_second_upper_estimate": 0.006545774612737742
     },
     {
       "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": 9370.218689687115,
-      "required_pages_per_second_upper_estimate": 0.0037379203325702552
+      "required_token_positions_per_second": 9439.844850718255,
+      "required_pages_per_second_upper_estimate": 0.003765695249209452
     },
     {
       "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": 18740.43737937423,
-      "required_pages_per_second_upper_estimate": 0.004335270367285118
+      "required_token_positions_per_second": 18879.68970143651,
+      "required_pages_per_second_upper_estimate": 0.0043674839412370615
     },
     {
       "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": 28110.656069061344,
-      "required_pages_per_second_upper_estimate": 0.0017117914693916994
+      "required_token_positions_per_second": 28319.534552154764,
+      "required_pages_per_second_upper_estimate": 0.0017245110731113837
     },
     {
       "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": 28110.656069061344,
-      "required_pages_per_second_upper_estimate": 0.0021640361194727406
+      "required_token_positions_per_second": 28319.534552154764,
+      "required_pages_per_second_upper_estimate": 0.0021801161633139207
     },
     {
       "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": 28110.656069061344,
-      "required_pages_per_second_upper_estimate": 0.34904026806388794
+      "required_token_positions_per_second": 28319.534552154764,
+      "required_pages_per_second_upper_estimate": 0.35163383975259527
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11812752755463407,
-    "producer_p95_seconds": 0.2938125309992756,
-    "gpu_update_seconds": 0.9835416125498919,
-    "producer_batches_per_second": 8.46542732842266,
-    "gpu_batches_per_second": 1.0167337987941747,
-    "producer_headroom": 8.326100045520747,
+    "producer_batch_seconds": 0.08179990309372442,
+    "producer_p95_seconds": 0.2153611630001251,
+    "gpu_update_seconds": 0.9762872320193671,
+    "producer_batches_per_second": 12.224953357881406,
+    "gpu_batches_per_second": 1.0242887207810607,
+    "producer_headroom": 11.935065875331905,
     "sustainable": true,
-    "buffer_seconds": 60.9795799780933,
+    "buffer_seconds": 60.52980838520076,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 453078.2982098044,
-    "upload_bytes_per_second": 5174292.444791185,
+    "generation_bytes_per_second": 472495.5792892196,
+    "upload_bytes_per_second": 5077822.97609272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.22005567284202,
+    "estimated_full_shards_per_hour": 101.38655218131485,
     "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": 1.157918292999966,
-      "bytes": 10292312,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 2.16527700000006,
       "bytes": 1181305,
@@ -562,6 +556,12 @@
       "bytes": 7035828,
       "pages": 21,
       "shards": 3
+    },
+    {
+      "seconds": 0.8532845239988092,
+      "bytes": 842129,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -150,3 +150,4 @@
 {"step": 170000, "loss": 0.5951601397246122, "lr": 0.0003, "grad_norm": 0.13240837884355422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24880314526, "tokens_per_second": 150478.4375103592, "data_wait_s": 1.1244999768678099e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 170100, "loss": 0.6039816215634346, "lr": 0.0003, "grad_norm": 0.13955668703862745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24895107560, "tokens_per_second": 135326.1570778702, "data_wait_s": 1.3150000086170621e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 170200, "loss": 0.5940169133245945, "lr": 0.0003, "grad_norm": 0.1530880732622695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24909840087, "tokens_per_second": 151088.30017296015, "data_wait_s": 1.4093000572756864e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 170300, "loss": 0.5909730903804302, "lr": 0.0003, "grad_norm": 0.14221754957390548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24924519187, "tokens_per_second": 149268.1498177246, "data_wait_s": 1.3880000551580451e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c5466e812fb428a86a4edbc7131850a797c348af2cfd14bc8744d4bdf07100d9
-size 324450
+oid sha256:86527d54d8d93856d6322f7e0793b9fe0916a7c0c1bd6a88f17a3d5098dec7c0
+size 326568