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": 76844.69940750617,
-      "required_pages_per_second_upper_estimate": 0.005316294609279436
+      "required_token_positions_per_second": 76977.65082288891,
+      "required_pages_per_second_upper_estimate": 0.0053254924966854
     },
     {
       "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": 9605.58742593827,
-      "required_pages_per_second_upper_estimate": 0.004021658867765338
+      "required_token_positions_per_second": 9622.206352861114,
+      "required_pages_per_second_upper_estimate": 0.004028616865425275
     },
     {
       "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": 19211.17485187654,
-      "required_pages_per_second_upper_estimate": 0.0040011810888954625
+      "required_token_positions_per_second": 19244.41270572223,
+      "required_pages_per_second_upper_estimate": 0.004008103657310368
     },
     {
       "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": 28816.76227781481,
-      "required_pages_per_second_upper_estimate": 0.0017275208979067199
+      "required_token_positions_per_second": 28866.619058583343,
+      "required_pages_per_second_upper_estimate": 0.001730509735786897
     },
     {
       "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": 28816.76227781481,
-      "required_pages_per_second_upper_estimate": 0.0022403765155583574
+      "required_token_positions_per_second": 28866.619058583343,
+      "required_pages_per_second_upper_estimate": 0.002244252660966308
     },
     {
       "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": 28816.76227781481,
-      "required_pages_per_second_upper_estimate": 0.3696304854711306
+      "required_token_positions_per_second": 28866.619058583343,
+      "required_pages_per_second_upper_estimate": 0.3702699947227889
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.23402521895314976,
-    "producer_p95_seconds": 0.3424311030030367,
-    "gpu_update_seconds": 0.959441582418348,
-    "producer_batches_per_second": 4.273043753460575,
-    "gpu_batches_per_second": 1.042272941182538,
-    "producer_headroom": 4.099735860563052,
+    "producer_batch_seconds": 0.12026568635951662,
+    "producer_p95_seconds": 0.3352762819995405,
+    "gpu_update_seconds": 0.9577844895479372,
+    "producer_batches_per_second": 8.314923651710986,
+    "gpu_batches_per_second": 1.0440762101628813,
+    "producer_headroom": 7.963904905384076,
     "sustainable": true,
-    "buffer_seconds": 60.44481969235592,
+    "buffer_seconds": 60.34042284152004,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370614.45573017484,
-    "upload_bytes_per_second": 2973634.2639670814,
+    "generation_bytes_per_second": 370039.57796102617,
+    "upload_bytes_per_second": 2963332.0061337235,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.5252347367185,
+    "estimated_full_shards_per_hour": 79.40187934992875,
     "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": 1.2937959129994852,
-      "bytes": 7777018,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.521280608001689,
       "bytes": 6127630,
@@ -562,6 +556,12 @@
       "bytes": 11546669,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.7511213550023967,
+      "bytes": 7640885,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -417,3 +417,4 @@
 {"step": 884200, "loss": 0.507845938205719, "lr": 3e-05, "grad_norm": 0.322197629398531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129421974009, "tokens_per_second": 151447.7346586818, "data_wait_s": 1.6657999367453158e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 884300, "loss": 0.5248423218727112, "lr": 3e-05, "grad_norm": 0.3125030448172315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129436576410, "tokens_per_second": 151040.53643126355, "data_wait_s": 1.5714998880866915e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 884400, "loss": 0.5472606420516968, "lr": 3e-05, "grad_norm": 0.30869917216185366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129451312048, "tokens_per_second": 152844.40345789696, "data_wait_s": 1.1632997484412044e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 884500, "loss": 0.5481231212615967, "lr": 3e-05, "grad_norm": 0.31447723963001145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129466019236, "tokens_per_second": 153058.77693741233, "data_wait_s": 1.3557997590396553e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:471fca93a95bd80bdfcf174826b65d7494eb98a04b69386d7c1500dee00443b1
-size 893871
+oid sha256:e6afa310fb3baba78094ff202600b02a1e3cad093a28c00d4a77bdea0adecea4
+size 895989