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": 76314.6956180984,
-      "required_pages_per_second_upper_estimate": 0.0066147643983490365
+      "required_token_positions_per_second": 75870.40987629912,
+      "required_pages_per_second_upper_estimate": 0.006576254836281793
     },
     {
       "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": 9539.3369522623,
-      "required_pages_per_second_upper_estimate": 0.0038053841360548507
+      "required_token_positions_per_second": 9483.80123453739,
+      "required_pages_per_second_upper_estimate": 0.00378323010792141
     },
     {
       "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": 19078.6739045246,
-      "required_pages_per_second_upper_estimate": 0.004413515434619125
+      "required_token_positions_per_second": 18967.60246907478,
+      "required_pages_per_second_upper_estimate": 0.004387821012818309
     },
     {
       "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": 28618.010856786896,
-      "required_pages_per_second_upper_estimate": 0.0017426867140793356
+      "required_token_positions_per_second": 28451.40370361217,
+      "required_pages_per_second_upper_estimate": 0.001732541212571172
     },
     {
       "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": 28618.010856786896,
-      "required_pages_per_second_upper_estimate": 0.0022030936954797946
+      "required_token_positions_per_second": 28451.40370361217,
+      "required_pages_per_second_upper_estimate": 0.0021902678156302865
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28618.010856786896,
-      "required_pages_per_second_upper_estimate": 0.35533991652019437
+      "required_token_positions_per_second": 28451.40370361217,
+      "required_pages_per_second_upper_estimate": 0.3532712132760367
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,30 +155,37 @@
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04698267994528749,
-    "producer_p95_seconds": 0.1392136389999905,
-    "gpu_update_seconds": 0.9661048819346277,
-    "producer_batches_per_second": 21.284439311774577,
-    "gpu_batches_per_second": 1.0350843047159612,
-    "producer_headroom": 20.563000728346726,
+    "producer_batch_seconds": 0.046732310742193306,
+    "producer_p95_seconds": 0.15708496400020522,
+    "gpu_update_seconds": 0.9717622472345654,
+    "producer_batches_per_second": 21.398471081746184,
+    "gpu_batches_per_second": 1.0290582936781023,
+    "producer_headroom": 20.79422634578153,
     "sustainable": true,
-    "buffer_seconds": 60.86460756188155,
+    "buffer_seconds": 60.249259328543054,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 360540.78813823167,
+    "upload_bytes_per_second": 21436.360429233762,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 77.36366017446721,
     "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": 1.0632401930001834,
+      "bytes": 22792,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -24,3 +24,4 @@
 {"step": 42400, "loss": 0.6546931639313698, "lr": 0.0003, "grad_norm": 0.13572722711902627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6204361023, "tokens_per_second": 150572.9859644851, "data_wait_s": 9.984999906009762e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 42500, "loss": 0.6561312675476074, "lr": 0.0003, "grad_norm": 0.13010470941836758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6218832513, "tokens_per_second": 148977.37935472347, "data_wait_s": 1.1410999832150992e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 42600, "loss": 0.8620638959109783, "lr": 0.0003, "grad_norm": 0.19228546236749122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6233543216, "tokens_per_second": 150894.77250834717, "data_wait_s": 1.3367000065045431e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 42700, "loss": 0.6080329138785601, "lr": 0.0003, "grad_norm": 0.12684101795425848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6248313748, "tokens_per_second": 150561.04743861072, "data_wait_s": 1.1466000159998657e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f310acb680f815acd151b56c9eb25d9dc1b90b2bc1c34ce16703dd91123dd82
-size 55233
+oid sha256:08df99a91ccb78ae3ad23be984ef0f2800f4cea4131e181b4773d6533e61852f
+size 57351