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": 74821.61194601047,
-      "required_pages_per_second_upper_estimate": 0.006485347689838425
+      "required_token_positions_per_second": 74492.57974304775,
+      "required_pages_per_second_upper_estimate": 0.0064568280124101115
     },
     {
       "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": 9352.701493251308,
-      "required_pages_per_second_upper_estimate": 0.0037309324610065855
+      "required_token_positions_per_second": 9311.57246788097,
+      "required_pages_per_second_upper_estimate": 0.0037145254778526288
     },
     {
       "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": 18705.402986502617,
-      "required_pages_per_second_upper_estimate": 0.004327165777078526
+      "required_token_positions_per_second": 18623.14493576194,
+      "required_pages_per_second_upper_estimate": 0.004308136824731918
     },
     {
       "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": 28058.104479753925,
-      "required_pages_per_second_upper_estimate": 0.0017085913533197582
+      "required_token_positions_per_second": 27934.717403642906,
+      "required_pages_per_second_upper_estimate": 0.0017010777277465547
     },
     {
       "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": 28058.104479753925,
-      "required_pages_per_second_upper_estimate": 0.002159990552655746
+      "required_token_positions_per_second": 27934.717403642906,
+      "required_pages_per_second_upper_estimate": 0.0021504918739794323
     },
     {
       "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": 28058.104479753925,
-      "required_pages_per_second_upper_estimate": 0.34838775320354526
+      "required_token_positions_per_second": 27934.717403642906,
+      "required_pages_per_second_upper_estimate": 0.34685569866822585
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11206334674176333,
-    "producer_p95_seconds": 0.3854381060009473,
-    "gpu_update_seconds": 0.9853837425101241,
-    "producer_batches_per_second": 8.923524319725889,
-    "gpu_batches_per_second": 1.0148330613336922,
-    "producer_headroom": 8.793095790551606,
+    "producer_batch_seconds": 0.26638985880470045,
+    "producer_p95_seconds": 0.3792188540028292,
+    "gpu_update_seconds": 0.989736162371003,
+    "producer_batches_per_second": 3.75389665540209,
+    "gpu_batches_per_second": 1.0103702764627787,
+    "producer_headroom": 3.715367269655008,
     "sustainable": true,
-    "buffer_seconds": 60.10840829311757,
+    "buffer_seconds": 60.37390590463119,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334421.5808160601,
-    "upload_bytes_per_second": 3209887.831880697,
+    "generation_bytes_per_second": 362517.05966757325,
+    "upload_bytes_per_second": 3220700.484804879,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.75908630715705,
+    "estimated_full_shards_per_hour": 77.78772203941725,
     "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": 1.1276710530000855,
-      "bytes": 5237494,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 0.9725549989998399,
       "bytes": 7624584,
@@ -562,6 +556,12 @@
       "bytes": 865291,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.3777121140010422,
+      "bytes": 7459243,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -323,3 +323,4 @@
 {"step": 347300, "loss": 0.5554037876427174, "lr": 0.0003, "grad_norm": 0.19573400654013476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50835576533, "tokens_per_second": 147477.54933544682, "data_wait_s": 1.2800002878066152e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 347400, "loss": 0.5926481354981661, "lr": 0.0003, "grad_norm": 0.1357957883806844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50850238243, "tokens_per_second": 149303.969908174, "data_wait_s": 8.74599936651066e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 347500, "loss": 0.5563265308737755, "lr": 0.0003, "grad_norm": 0.14774604212781411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50864859371, "tokens_per_second": 148031.7105087188, "data_wait_s": 1.746600173646584e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 347600, "loss": 0.600357124581933, "lr": 0.0003, "grad_norm": 0.16409013175819187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50879520528, "tokens_per_second": 150538.73478444447, "data_wait_s": 1.3614000636152923e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0a93b5b18500414dedae74f387b75090868c0dc9f83795f1dc8ec17a205ba984
-size 693213
+oid sha256:7776d8fb366bd3443f1554638f89fe39fe0051ba592a22dc7c2fcb50e023c870
+size 695331