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": 75255.1226721041,
-      "required_pages_per_second_upper_estimate": 0.006522923300853331
+      "required_token_positions_per_second": 74165.07925041376,
+      "required_pages_per_second_upper_estimate": 0.006428441099751007
     },
     {
       "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": 9406.890334013013,
-      "required_pages_per_second_upper_estimate": 0.0037525491997817987
+      "required_token_positions_per_second": 9270.63490630172,
+      "required_pages_per_second_upper_estimate": 0.0036981948724675766
     },
     {
       "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": 18813.780668026026,
-      "required_pages_per_second_upper_estimate": 0.004352237046316917
+      "required_token_positions_per_second": 18541.26981260344,
+      "required_pages_per_second_upper_estimate": 0.004289196455942152
     },
     {
       "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": 28220.67100203904,
-      "required_pages_per_second_upper_estimate": 0.0017184908016062964
+      "required_token_positions_per_second": 27811.904718905163,
+      "required_pages_per_second_upper_estimate": 0.0016935990795943852
     },
     {
       "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": 28220.67100203904,
-      "required_pages_per_second_upper_estimate": 0.0021725053735541897
+      "required_token_positions_per_second": 27811.904718905163,
+      "required_pages_per_second_upper_estimate": 0.0021410374135410505
     },
     {
       "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": 28220.67100203904,
-      "required_pages_per_second_upper_estimate": 0.3504062853351756
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27811.904718905163,
+      "required_pages_per_second_upper_estimate": 0.3453307761513983
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.13972836331254257,
-    "producer_p95_seconds": 0.26482249600121577,
-    "gpu_update_seconds": 0.9797073924288455,
-    "producer_batches_per_second": 7.156743099918899,
-    "gpu_batches_per_second": 1.0207129268677313,
-    "producer_headroom": 7.0115141207046765,
+    "producer_batch_seconds": 0.42058296425783226,
+    "producer_p95_seconds": 0.2706223980003415,
+    "gpu_update_seconds": 0.9941066704865508,
+    "producer_batches_per_second": 2.3776521756286937,
+    "gpu_batches_per_second": 1.0059282667428082,
+    "producer_headroom": 2.3636398878893443,
     "sustainable": true,
-    "buffer_seconds": 60.74185833058842,
+    "buffer_seconds": 60.6405068996796,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 427844.63215868216,
-    "upload_bytes_per_second": 5289281.255193045,
+    "generation_bytes_per_second": 389024.0675595172,
+    "upload_bytes_per_second": 5334482.115037312,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.8054983479533,
+    "estimated_full_shards_per_hour": 83.47550888146532,
     "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": 1.0039940289998412,
-      "bytes": 2541534,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.7366310759998669,
       "bytes": 10198556,
@@ -562,6 +556,12 @@
       "bytes": 5823579,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.1941432809999242,
+      "bytes": 7137236,
+      "pages": 22,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -135,3 +135,4 @@
 {"step": 208500, "loss": 0.6042758673429489, "lr": 0.0003, "grad_norm": 0.14450466836031053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30516426206, "tokens_per_second": 148336.9974279627, "data_wait_s": 2.116799987561535e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 208600, "loss": 0.6167097222059965, "lr": 0.0003, "grad_norm": 0.14051531890274305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30531144904, "tokens_per_second": 149534.9880224501, "data_wait_s": 2.0044000848429278e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 208700, "loss": 0.5843361467123032, "lr": 0.0003, "grad_norm": 0.1446629908582026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30545964578, "tokens_per_second": 142373.67929929346, "data_wait_s": 1.7577000107849017e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 208800, "loss": 0.64070232398808, "lr": 0.0003, "grad_norm": 0.1767233183359933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30560615198, "tokens_per_second": 148293.30779086935, "data_wait_s": 1.477899968449492e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04051d6a2a3b2d81708d5e8a66e301789db0d28b40d68a3a37397e030d7a35eb
-size 291897
+oid sha256:389bd8ce94e390011c2a6388d42557705b21a962d6b32f478023d5909d4f0fbf
+size 294015