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": 74340.72972774111,
-      "required_pages_per_second_upper_estimate": 0.006443666037943677
+      "required_token_positions_per_second": 74700.12224828225,
+      "required_pages_per_second_upper_estimate": 0.006474817270752167
     },
     {
       "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": 9292.591215967639,
-      "required_pages_per_second_upper_estimate": 0.003706953572669395
+      "required_token_positions_per_second": 9337.515281035281,
+      "required_pages_per_second_upper_estimate": 0.0037248744538995056
     },
     {
       "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": 18585.182431935278,
-      "required_pages_per_second_upper_estimate": 0.004299354867588868
+      "required_token_positions_per_second": 18675.030562070562,
+      "required_pages_per_second_upper_estimate": 0.0043201396512225826
     },
     {
       "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": 27877.773647902915,
-      "required_pages_per_second_upper_estimate": 0.0016976101517827862
+      "required_token_positions_per_second": 28012.545843105843,
+      "required_pages_per_second_upper_estimate": 0.0017058170713755835
     },
     {
       "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": 27877.773647902915,
-      "required_pages_per_second_upper_estimate": 0.002146108186032163
+      "required_token_positions_per_second": 28012.545843105843,
+      "required_pages_per_second_upper_estimate": 0.0021564833226921983
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27877.773647902915,
-      "required_pages_per_second_upper_estimate": 0.3461486478004261
+      "required_token_positions_per_second": 28012.545843105843,
+      "required_pages_per_second_upper_estimate": 0.34782206741132454
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.20213575359375113,
-    "producer_p95_seconds": 0.17468009399999573,
-    "gpu_update_seconds": 0.9917578193006026,
-    "producer_batches_per_second": 4.947170316092532,
-    "gpu_batches_per_second": 1.008310678815933,
-    "producer_headroom": 4.906394844396602,
+    "producer_batch_seconds": 0.06972003900783008,
+    "producer_p95_seconds": 0.16806734299962045,
+    "gpu_update_seconds": 0.9869863365812015,
+    "producer_batches_per_second": 14.34307860739568,
+    "gpu_batches_per_second": 1.0131852518484463,
+    "producer_headroom": 14.156422610009663,
     "sustainable": true,
-    "buffer_seconds": 60.49722697733676,
+    "buffer_seconds": 60.20616653145329,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 168,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 587684.2498395673,
+    "generation_bytes_per_second": 568654.3821032892,
     "upload_bytes_per_second": 5665389.843365814,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.05167971320279,
+    "estimated_full_shards_per_hour": 61.009996401424445,
     "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": [

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

@@ -31,3 +31,4 @@
 {"step": 85000, "loss": 0.6123912185430527, "lr": 0.0003, "grad_norm": 0.12991575957755994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12439396796, "tokens_per_second": 148490.2662518952, "data_wait_s": 1.2061000234098174e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 85100, "loss": 0.6527527756989002, "lr": 0.0003, "grad_norm": 0.1299015585553546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12454100277, "tokens_per_second": 134800.67630284632, "data_wait_s": 9.854999916569795e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 85200, "loss": 0.602685734629631, "lr": 0.0003, "grad_norm": 0.13874306371124315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12468770344, "tokens_per_second": 148680.2669662592, "data_wait_s": 1.0623999969539e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 85300, "loss": 0.6185211539268494, "lr": 0.0003, "grad_norm": 0.1325468472825569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12483365755, "tokens_per_second": 147805.61911232604, "data_wait_s": 2.027199980147998e-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 sha256:01f41aff2ce7ba40d890aa799299aba352e99f4c2dbe9f4341ee2d637e778200
-size 70842
+oid sha256:937aebc5fd556bd2ca1b4f9c2a336d70e26a7c426be2492188bd95e2242f4afc
+size 72960