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": 5801.295609408763,
-      "required_pages_per_second_upper_estimate": 0.0005028416001742531
+      "required_token_positions_per_second": 71392.83449990141,
+      "required_pages_per_second_upper_estimate": 0.00618814995096668
     },
     {
       "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": 725.1619511760954,
-      "required_pages_per_second_upper_estimate": 0.0002892779444615029
+      "required_token_positions_per_second": 8924.104312487676,
+      "required_pages_per_second_upper_estimate": 0.0035599586375010673
     },
     {
       "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": 1450.3239023521908,
-      "required_pages_per_second_upper_estimate": 0.00033550690998027907
+      "required_token_positions_per_second": 17848.208624975352,
+      "required_pages_per_second_upper_estimate": 0.004128868947644701
     },
     {
       "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": 2175.485853528286,
-      "required_pages_per_second_upper_estimate": 0.00013247567458771122
+      "required_token_positions_per_second": 26772.31293746303,
+      "required_pages_per_second_upper_estimate": 0.0016302933944211046
     },
     {
       "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": 2175.485853528286,
-      "required_pages_per_second_upper_estimate": 0.0001674749231348835
+      "required_token_positions_per_second": 26772.31293746303,
+      "required_pages_per_second_upper_estimate": 0.002061006760431398
     },
     {
       "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": 2,
-      "required_token_positions_per_second": 2175.485853528286,
-      "required_pages_per_second_upper_estimate": 0.027012253418035016
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 26772.31293746303,
+      "required_pages_per_second_upper_estimate": 0.3324225255157633
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 5,
+    "prefetch_batches": 59,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05747767319047806,
+    "producer_batch_seconds": 0.055598538142860005,
     "producer_p95_seconds": 0.14835468800004037,
-    "gpu_update_seconds": 12.708885215300029,
-    "producer_batches_per_second": 17.39805988120033,
-    "gpu_batches_per_second": 0.07868510754948951,
-    "producer_headroom": 221.10994599909145,
+    "gpu_update_seconds": 1.032708681710933,
+    "producer_batches_per_second": 17.98608440802001,
+    "gpu_batches_per_second": 0.9683272908515274,
+    "producer_headroom": 18.57438551814791,
     "sustainable": true,
-    "buffer_seconds": 63.544426076500145,
+    "buffer_seconds": 60.929812220945045,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 17,
+    "upload_pages_per_commit": 24,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -1 +1,2 @@
 {"step": 155100, "loss": 0.5673591010272503, "lr": 0.0003, "grad_norm": 0.13793740573041358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22698961668, "tokens_per_second": 44241.941022346604, "data_wait_s": 1.6559999949095072e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 155200, "loss": 0.6261652167886496, "lr": 0.0003, "grad_norm": 0.146581492453678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22713619019, "tokens_per_second": 150427.58678988548, "data_wait_s": 1.6635999941172486e-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 sha256:18b8e1a8a7ba0cc74041f78ee6898e8f0e38f92b7e414bae3c4649ec295683f8
-size 2283
+oid sha256:be4d4553846fe771524261c2a2353abb88f3d24e12ea5fde7b47e1441873c82b
+size 4401