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": 77286.43891436518,
-      "required_pages_per_second_upper_estimate": 0.006698992644408573
+      "required_token_positions_per_second": 6402.112457587517,
+      "required_pages_per_second_upper_estimate": 0.0005549188818180076
     },
     {
       "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": 9660.804864295647,
-      "required_pages_per_second_upper_estimate": 0.003853839502272079
+      "required_token_positions_per_second": 800.2640571984397,
+      "required_pages_per_second_upper_estimate": 0.00031923729743036527
     },
     {
       "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": 19321.609728591295,
-      "required_pages_per_second_upper_estimate": 0.004469714361992479
+      "required_token_positions_per_second": 1600.5281143968793,
+      "required_pages_per_second_upper_estimate": 0.00037025401093297266
     },
     {
       "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": 28982.414592886944,
-      "required_pages_per_second_upper_estimate": 0.001764876989729183
+      "required_token_positions_per_second": 2400.792171595319,
+      "required_pages_per_second_upper_estimate": 0.00014619564726709943
     },
     {
       "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": 28982.414592886944,
-      "required_pages_per_second_upper_estimate": 0.0022311465038188797
+      "required_token_positions_per_second": 2400.792171595319,
+      "required_pages_per_second_upper_estimate": 0.0001848196271874899
     },
     {
       "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": 28982.414592886944,
-      "required_pages_per_second_upper_estimate": 0.3598645913417056
+      "estimated_pages_for_buffer": 3,
+      "required_token_positions_per_second": 2400.792171595319,
+      "required_pages_per_second_upper_estimate": 0.02980980383668772
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,30 +155,43 @@
   "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": 63,
+    "prefetch_batches": 6,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 1.3053112965000082,
-    "producer_p95_seconds": 2.5095109330000014,
-    "gpu_update_seconds": 0.9539577840000106,
-    "producer_batches_per_second": 0.7661007781678948,
-    "gpu_batches_per_second": 1.0482644166987465,
-    "producer_headroom": 0.7308278006617287,
-    "sustainable": false,
-    "buffer_seconds": 60.09934039200067,
-    "buffer_drain_seconds": 223.2746938195993,
+    "producer_batch_seconds": 0.04666193766666091,
+    "producer_p95_seconds": 0.12250736300001108,
+    "gpu_update_seconds": 11.516198831000015,
+    "producer_batches_per_second": 21.430743128236646,
+    "gpu_batches_per_second": 0.0868342075953168,
+    "producer_headroom": 246.80069896086047,
+    "sustainable": true,
+    "buffer_seconds": 69.0971929860001,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 7,
-    "max_pending_pages": 1713,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "upload_pages_per_commit": 21,
+    "max_pending_pages": 4096,
+    "generation_bytes_per_second": 381132.484969901,
+    "upload_bytes_per_second": 159497.9575335065,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 81.78215896437428,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 1.104392732000008,
+      "bytes": 36786,
+      "pages": 1,
+      "shards": 1
+    },
+    {
+      "seconds": 0.71818233099998,
+      "bytes": 253911,
+      "pages": 7,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -0,0 +1 @@
+{"step": 82000, "loss": 0.6297517661005259, "lr": 0.0003, "grad_norm": 0.13215398907681755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11999942244, "tokens_per_second": 40155.054755347504, "data_wait_s": 1.227600000675011e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:988ee53fd1cd660f7d9c599e165c7ad11354d6a46f576235186c6b5db5eb77de
-size 165
+oid sha256:977a7a149c1ff7f972c63714b1482fd8e196081bdcaf994a011d9a1f86c59d15
+size 2283