Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77131.3897560337,
-      "required_pages_per_second_upper_estimate": 0.005336128512803829
+      "required_token_positions_per_second": 76757.56266358808,
+      "required_pages_per_second_upper_estimate": 0.005310266287150081
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9641.423719504213,
-      "required_pages_per_second_upper_estimate": 0.0040366627755344915
+      "required_token_positions_per_second": 9594.69533294851,
+      "required_pages_per_second_upper_estimate": 0.004017098575883291
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19282.847439008427,
-      "required_pages_per_second_upper_estimate": 0.004016108598661806
+      "required_token_positions_per_second": 19189.39066589702,
+      "required_pages_per_second_upper_estimate": 0.0039966440174435455
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28924.27115851264,
-      "required_pages_per_second_upper_estimate": 0.001733965891147999
+      "required_token_positions_per_second": 28784.08599884553,
+      "required_pages_per_second_upper_estimate": 0.0017255620048763005
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28924.27115851264,
-      "required_pages_per_second_upper_estimate": 0.002248734858151023
+      "required_token_positions_per_second": 28784.08599884553,
+      "required_pages_per_second_upper_estimate": 0.0022378360785962575
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28924.27115851264,
-      "required_pages_per_second_upper_estimate": 0.3710094939586799
+      "required_token_positions_per_second": 28784.08599884553,
+      "required_pages_per_second_upper_estimate": 0.3692113492495675
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.18135378182036987,
-    "producer_p95_seconds": 0.15660588199989434,
-    "gpu_update_seconds": 0.9558754254681704,
-    "producer_batches_per_second": 5.514084073474109,
-    "gpu_batches_per_second": 1.0461614278975926,
-    "producer_headroom": 5.270777459799326,
+    "producer_batch_seconds": 0.10828829944531293,
+    "producer_p95_seconds": 0.16543441299927508,
+    "gpu_update_seconds": 0.9605307599869214,
+    "producer_batches_per_second": 9.234608033576274,
+    "gpu_batches_per_second": 1.0410910734536143,
+    "producer_headroom": 8.870125072672348,
     "sustainable": true,
-    "buffer_seconds": 60.22015180449473,
+    "buffer_seconds": 60.513437879176045,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 152,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 573754.1152193514,
-    "upload_bytes_per_second": 3371229.1688682893,
+    "generation_bytes_per_second": 532768.701208936,
+    "upload_bytes_per_second": 3529238.6212974694,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.55713840692237,
+    "estimated_full_shards_per_hour": 114.31976105881749,
     "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": [
@@ -196,6 +196,12 @@
       "bytes": 6805061,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.581821284000398,
+      "bytes": 6273764,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -42,3 +42,4 @@
 {"step": 929200, "loss": 0.5404407978057861, "lr": 3e-05, "grad_norm": 0.34774026171496636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136008349606, "tokens_per_second": 153824.6512763019, "data_wait_s": 1.460599924030248e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 929300, "loss": 0.5592553615570068, "lr": 3e-05, "grad_norm": 0.3482657231766738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136023121458, "tokens_per_second": 153385.71312137245, "data_wait_s": 1.420599983248394e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 929400, "loss": 0.5707401633262634, "lr": 3e-05, "grad_norm": 0.3463429130595973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136037969906, "tokens_per_second": 154595.82418027864, "data_wait_s": 9.903000318445265e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 929500, "loss": 0.558310329914093, "lr": 3e-05, "grad_norm": 0.3346028068580802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136052657431, "tokens_per_second": 153483.4047636319, "data_wait_s": 9.879000572254881e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c410e0301bbe8ba69aa1a8fc57da52abf6cd24d201f47d5e10ab20135c7fc6b2
-size 93357
+oid sha256:293647a1ac9e83cec8c408a2deab1de9c18cfc8e7cf6e48b1b00d881ae7baad4
+size 95475