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": 76548.08554960876,
-      "required_pages_per_second_upper_estimate": 0.005295774174351123
+      "required_token_positions_per_second": 76532.75640277524,
+      "required_pages_per_second_upper_estimate": 0.005294713668404658
     },
     {
       "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": 9568.510693701095,
-      "required_pages_per_second_upper_estimate": 0.004006135614227845
+      "required_token_positions_per_second": 9566.594550346905,
+      "required_pages_per_second_upper_estimate": 0.004005333365019069
     },
     {
       "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": 19137.02138740219,
-      "required_pages_per_second_upper_estimate": 0.003985736877803819
+      "required_token_positions_per_second": 19133.18910069381,
+      "required_pages_per_second_upper_estimate": 0.0039849387135466605
     },
     {
       "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": 28705.532081103283,
-      "required_pages_per_second_upper_estimate": 0.0017208528174525403
+      "required_token_positions_per_second": 28699.783651040718,
+      "required_pages_per_second_upper_estimate": 0.0017205082078476338
     },
     {
       "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": 28705.532081103283,
-      "required_pages_per_second_upper_estimate": 0.002231728857013966
+      "required_token_positions_per_second": 28699.783651040718,
+      "required_pages_per_second_upper_estimate": 0.002231281941861273
     },
     {
       "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": 28705.532081103283,
-      "required_pages_per_second_upper_estimate": 0.3682037439373954
+      "required_token_positions_per_second": 28699.783651040718,
+      "required_pages_per_second_upper_estimate": 0.3681300092487361
     }
   ],
   "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": 0.08237858583609636,
-    "producer_p95_seconds": 0.2233849910007848,
-    "gpu_update_seconds": 0.9631592935426042,
-    "producer_batches_per_second": 12.139077041084912,
-    "gpu_batches_per_second": 1.0382498582574973,
-    "producer_headroom": 11.691864867150588,
+    "producer_batch_seconds": 0.12087911760932002,
+    "producer_p95_seconds": 0.2343866369992611,
+    "gpu_update_seconds": 0.9633522097647389,
+    "producer_batches_per_second": 8.272727496505965,
+    "gpu_batches_per_second": 1.0380419433970167,
+    "producer_headroom": 7.969550314540537,
     "sustainable": true,
-    "buffer_seconds": 60.67903549318407,
+    "buffer_seconds": 60.691189215178554,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285198.63712742296,
-    "upload_bytes_per_second": 4539089.648220684,
+    "generation_bytes_per_second": 297577.1925434661,
+    "upload_bytes_per_second": 4414033.870093756,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.19698844306008,
+    "estimated_full_shards_per_hour": 63.853138277320745,
     "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": 1.1612172679997457,
-      "bytes": 6026166,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.3405283600004623,
       "bytes": 7840338,
@@ -562,6 +556,12 @@
       "bytes": 16596361,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 3.107019715000206,
+      "bytes": 2583987,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -191,3 +191,4 @@
 {"step": 459300, "loss": 0.5798925757408142, "lr": 0.0003, "grad_norm": 0.1686032957160167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67227281839, "tokens_per_second": 151925.63496468656, "data_wait_s": 8.195998816518113e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 459400, "loss": 0.5622214078903198, "lr": 0.0003, "grad_norm": 0.16292578632419594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67241918117, "tokens_per_second": 122496.68243172229, "data_wait_s": 1.2632001016754657e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 459500, "loss": 0.591704785823822, "lr": 0.0003, "grad_norm": 0.18903362584495223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67256612063, "tokens_per_second": 152532.32364607448, "data_wait_s": 1.0562998795649037e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 459600, "loss": 0.6140382289886475, "lr": 0.0003, "grad_norm": 0.1880881308400727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67271305849, "tokens_per_second": 153336.56193547978, "data_wait_s": 1.7261001630686224e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5fbc429a6cead5f2e043d2d87cfd4c69b9a01593b08a54bd33a4b5827b7a4ba7
-size 411288
+oid sha256:a405c950dd17d7fc0e544ad88abcba3dd78f9cc6c2475b196e0342597ccf208b
+size 413406