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": 75727.89197689154,
-      "required_pages_per_second_upper_estimate": 0.006563901745969458
+      "required_token_positions_per_second": 74388.34369265652,
+      "required_pages_per_second_upper_estimate": 0.006447793095746054
     },
     {
       "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": 9465.986497111442,
-      "required_pages_per_second_upper_estimate": 0.003776123542808139
+      "required_token_positions_per_second": 9298.542961582065,
+      "required_pages_per_second_upper_estimate": 0.0037093278129815164
     },
     {
       "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": 18931.972994222884,
-      "required_pages_per_second_upper_estimate": 0.0043795787608686635
+      "required_token_positions_per_second": 18597.08592316413,
+      "required_pages_per_second_upper_estimate": 0.004302108530790372
     },
     {
       "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": 28397.959491334324,
-      "required_pages_per_second_upper_estimate": 0.0017292867404435626
+      "required_token_positions_per_second": 27895.628884746195,
+      "required_pages_per_second_upper_estimate": 0.001698697442027354
     },
     {
       "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": 28397.959491334324,
-      "required_pages_per_second_upper_estimate": 0.0021861535322260893
+      "required_token_positions_per_second": 27895.628884746195,
+      "required_pages_per_second_upper_estimate": 0.002147482731590818
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28397.959491334324,
-      "required_pages_per_second_upper_estimate": 0.3526076150258183
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27895.628884746195,
+      "required_pages_per_second_upper_estimate": 0.346370350084386
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07430449135932804,
-    "producer_p95_seconds": 0.20834141100021952,
-    "gpu_update_seconds": 0.9735910782053486,
-    "producer_batches_per_second": 13.45813667122912,
-    "gpu_batches_per_second": 1.0271252709539325,
-    "producer_headroom": 13.1027217923769,
+    "producer_batch_seconds": 0.34071526806246766,
+    "producer_p95_seconds": 0.1939346020008088,
+    "gpu_update_seconds": 0.9911230219698828,
+    "producer_batches_per_second": 2.93500201997598,
+    "gpu_batches_per_second": 1.008956484546665,
+    "producer_headroom": 2.908948071526303,
     "sustainable": true,
-    "buffer_seconds": 60.362646848731615,
+    "buffer_seconds": 60.45850434016285,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380972.03719884646,
-    "upload_bytes_per_second": 5889066.857021001,
+    "generation_bytes_per_second": 347259.9973041408,
+    "upload_bytes_per_second": 5817932.865560684,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.74773060773893,
+    "estimated_full_shards_per_hour": 74.5139116224591,
     "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": [
@@ -472,6 +472,12 @@
       "bytes": 5998979,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.7888650649983902,
+      "bytes": 730238,
+      "pages": 23,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -84,3 +84,4 @@
 {"step": 48400, "loss": 0.6108626779168844, "lr": 0.0003, "grad_norm": 0.13623167526279106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7082980437, "tokens_per_second": 148360.78684643866, "data_wait_s": 1.2336999134276994e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 48500, "loss": 0.6089311204850674, "lr": 0.0003, "grad_norm": 0.12629009653325401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7097585312, "tokens_per_second": 149600.68584728957, "data_wait_s": 1.1580999853322282e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 48600, "loss": 0.6288862973451614, "lr": 0.0003, "grad_norm": 0.13536074438067727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7112279934, "tokens_per_second": 150000.63836483276, "data_wait_s": 9.921999662765302e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 48700, "loss": 0.6577148996293545, "lr": 0.0003, "grad_norm": 0.13764597902065967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7126931924, "tokens_per_second": 149801.73241544084, "data_wait_s": 1.3361001037992537e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:58f01db79ee34c3f386e05ccdd0b0661cca2da33ff66336a07b30f778a7bf310
-size 183096
+oid sha256:b3ab2f8dcfb7ba4ac372456e308131340e931068f0edfe5c14bd591ac7bdca1d
+size 185214