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": 75897.34787940634,
-      "required_pages_per_second_upper_estimate": 0.006578589754117382
+      "required_token_positions_per_second": 74569.67148891005,
+      "required_pages_per_second_upper_estimate": 0.006463510129554212
     },
     {
       "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": 9487.168484925793,
-      "required_pages_per_second_upper_estimate": 0.0037845733544462235
+      "required_token_positions_per_second": 9321.208936113757,
+      "required_pages_per_second_upper_estimate": 0.003718369609108727
     },
     {
       "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": 18974.336969851585,
-      "required_pages_per_second_upper_estimate": 0.004389378921049857
+      "required_token_positions_per_second": 18642.417872227514,
+      "required_pages_per_second_upper_estimate": 0.0043125952793911316
     },
     {
       "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": 28461.505454777376,
-      "required_pages_per_second_upper_estimate": 0.0017331563562173406
+      "required_token_positions_per_second": 27963.62680834127,
+      "required_pages_per_second_upper_estimate": 0.0017028381588167607
     },
     {
       "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": 28461.505454777376,
-      "required_pages_per_second_upper_estimate": 0.002191045476398421
+      "required_token_positions_per_second": 27963.62680834127,
+      "required_pages_per_second_upper_estimate": 0.0021527173999794664
     },
     {
       "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": 28461.505454777376,
-      "required_pages_per_second_upper_estimate": 0.3533966432171223
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27963.62680834127,
+      "required_pages_per_second_upper_estimate": 0.34721465672102597
     }
   ],
   "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": 0.12989549892171226,
-    "producer_p95_seconds": 0.5872306089950143,
-    "gpu_update_seconds": 0.9714173427660051,
-    "producer_batches_per_second": 7.698496162693812,
-    "gpu_batches_per_second": 1.0294236637289271,
-    "producer_headroom": 7.47845268565831,
+    "producer_batch_seconds": 0.32403751150013704,
+    "producer_p95_seconds": 0.3914885769991088,
+    "gpu_update_seconds": 0.9887129516316131,
+    "producer_batches_per_second": 3.0860624603938085,
+    "gpu_batches_per_second": 1.0114159001859544,
+    "producer_headroom": 3.0512299241354808,
     "sustainable": true,
-    "buffer_seconds": 60.227875251492314,
+    "buffer_seconds": 60.3114900495284,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363180.0047451357,
-    "upload_bytes_per_second": 4833264.740068988,
+    "generation_bytes_per_second": 364097.5403122029,
+    "upload_bytes_per_second": 4798137.863036595,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.92997462049058,
+    "estimated_full_shards_per_hour": 78.12685639404836,
     "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": 1.2379260390007403,
-      "bytes": 10087351,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.0402059740008553,
       "bytes": 6293747,
@@ -562,6 +556,12 @@
       "bytes": 7143913,
       "pages": 21,
       "shards": 3
+    },
+    {
+      "seconds": 1.1386764790004236,
+      "bytes": 6532415,
+      "pages": 23,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -400,3 +400,4 @@
 {"step": 195000, "loss": 0.587149441242218, "lr": 0.0003, "grad_norm": 0.14609337480068943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28540343020, "tokens_per_second": 150961.18382806727, "data_wait_s": 1.727200287859887e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 195100, "loss": 0.5759536698460579, "lr": 0.0003, "grad_norm": 0.1433815500195194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28555098636, "tokens_per_second": 135442.66935053776, "data_wait_s": 9.759001841302961e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 195200, "loss": 0.611761312931776, "lr": 0.0003, "grad_norm": 0.13731755191137826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28569686874, "tokens_per_second": 148886.4980394214, "data_wait_s": 1.1459997040219605e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 195300, "loss": 0.5979965794831514, "lr": 0.0003, "grad_norm": 0.13250483478838535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28584231055, "tokens_per_second": 149748.62037063212, "data_wait_s": 1.3190001482143998e-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 sha256:957b562bc43e3dad74bc39305131f9d30c94b3c37c9db7b9ec3987a7531190bf
-size 857865
+oid sha256:19fb419959154dc735c4c3bd9c500f6593ee82afb9323ce254b1b61cf3bfb40a
+size 859983